spybyscript commited on about 7 hours ago

Commit

d933c90

verified ·

1 Parent(s): 0c44c02

Scylla's Band v2-20261020

Browse files

This view is limited to 50 files because it contains too many changes. See raw diff

Files changed (50) hide show

README.md +10 -9
SHA256SUMS.json +188 -296
THIRD_PARTY_NOTICES.md +1 -1
coreai/coreai/duration_predictor.aimodel/main.hash +0 -1
coreai/coreai/duration_predictor.aimodel/main.mlirb +0 -3
coreai/coreai/duration_predictor.aimodel/metadata.json +0 -5
coreai/coreai/g2p.aimodel/main.hash +0 -0
coreai/coreai/g2p.aimodel/metadata.json +0 -5
coreai/coreai/vector_context_encoder.aimodel/main.hash +0 -0
coreai/coreai/vector_context_encoder.aimodel/metadata.json +0 -5
coreai/coreai/vector_estimator.aimodel/main.hash +0 -1
coreai/coreai/vector_estimator.aimodel/metadata.json +0 -5
coreai/coreai/vocoder.aimodel/main.hash +0 -1
coreai/coreai/vocoder.aimodel/main.mlirb +0 -3
coreai/coreai/vocoder.aimodel/metadata.json +0 -5
coreml/assets/fixed_references.json +0 -0
coreml/assets/g2p/export_config.json +0 -33
coreml/assets/g2p/language_map.json +0 -47
coreml/assets/g2p/normalization.json +0 -40
coreml/assets/g2p/pronunciation_overrides.json +0 -4
coreml/assets/languages.json +0 -30
coreml/assets/phone_vocab.json +0 -111
coreml/assets/voices.json +0 -42
coreml/coreml/duration_predictor.mlmodelc/model.mil +0 -0
coreml/coreml/duration_predictor.mlmodelc/weights/weight.bin +0 -3
coreml/coreml/g2p.mlmodelc/model.mil +0 -0
coreml/coreml/g2p.mlmodelc/weights/weight.bin +0 -3
coreml/coreml/vector_context_encoder.mlmodelc/analytics/coremldata.bin +0 -3
coreml/coreml/vector_context_encoder.mlmodelc/coremldata.bin +0 -3
coreml/coreml/vector_context_encoder.mlmodelc/model.mil +0 -267
coreml/coreml/vector_context_encoder.mlmodelc/weights/weight.bin +0 -3
coreml/coreml/vector_estimator.mlmodelc/analytics/coremldata.bin +0 -3
coreml/coreml/vector_estimator.mlmodelc/coremldata.bin +0 -3
coreml/coreml/vector_estimator.mlmodelc/model.mil +0 -0
coreml/coreml/vector_estimator.mlmodelc/weights/weight.bin +0 -3
coreml/coreml/vocoder.mlmodelc/analytics/coremldata.bin +0 -3
coreml/coreml/vocoder.mlmodelc/coremldata.bin +0 -3
coreml/coreml/vocoder.mlmodelc/model.mil +0 -0
coreml/coreml/vocoder.mlmodelc/weights/weight.bin +0 -3
coreml/manifest.json +0 -998
litert/assets/fixed_references.json +0 -0
litert/assets/voices.json +16 -0
litert/litert/duration_predictor.tflite +2 -2
litert/litert/vector_context_encoder.tflite +2 -2
litert/litert/vector_estimator.tflite +2 -2
litert/litert/vocoder.tflite +2 -2
litert/manifest.json +107 -7
onnx/assets/fixed_references.json +0 -0
onnx/assets/voices.json +16 -0
onnx/manifest.json +110 -10

README.md CHANGED Viewed

@@ -8,7 +8,7 @@ tags: [text-to-speech, tts, litert, onnx, coreml, coreai, pytorch, multilingual,
 # Scylla's Band v2
-Release **`v2-20261006`**. The [Scylla's Band runtime](https://github.com/lowkeytea/scyllasband) supports this release. Tag `v2-20261006-apple` adds Core ML and Core AI bundles of the same model; the runtime downloads from that tag.
 The model exposes **energy, tension, valence and assertiveness** on **0–4 scales with neutral at 2**. Combining the controls changes the generated delivery; a request is not a guarantee of a particular perceived emotion or exact external scorer value.
@@ -55,7 +55,7 @@ Combining the controls gives familiar deliveries:
 | Angry | `energy=2.3,tension=2.4,valence=1.4,assertiveness=2.4` |
 | Sad | `energy=1.7,tension=1.5,valence=1.4,assertiveness=2` |
-These are moderate styles; how strongly each comes through varies with voice, language and text. The [voice gallery](https://lowkeytea.github.io/scyllasband/#delivery) plays each of them for all ten voices.
 `--delivery auto` omits all coordinates. A per-axis `auto` omits just that coordinate; other unspecified fields in a partial request use neutral defaults. Omission is represented by a separate mask, not by the value 2. It is not an automatic text-to-performance planner.
@@ -72,7 +72,7 @@ Controls are **global within each request**. This release does not support contr
 | Audio | Mono 24 kHz; 100 mel features; 24 latent channels |
 | Mel / latent hop | 256 / 512 samples |
 | Duration model | Hidden size 320; 8 layers; reads the neighbouring text as context |
-| Vector model | Hidden size 512; 12 layers; adaptive layer normalization and QK normalization |
 | Maximum graph phones per sentence | 512 |
 | Trained target length | 64–420 latent frames (about 1.4–9 s) per pass |
 | Latent buckets | 128, 256, 384, 512, 768, 1024 frames |
@@ -114,16 +114,17 @@ The GPU is the default on Apple platforms. The Neural Engine placement runs the
 ## Voices and language coverage
-Ten voices: Ariadne, Felix, Gwen, Ink, Max, Orpheus, Rex, Scylla, Stone, Tuesday.
-All ten have Spanish (`es`), Italian (`it`), French (`fr`), German (`de`) and Vietnamese (`vi`). Ink, Orpheus and Tuesday use British English (`en_gb`); the other seven use American English (`en_us`). This is **60 trained voice/locale pairs**, not every combination of ten voices and seven locale IDs.
 ## Artifacts
-- `litert/` (164 MB): LiteRT bundle; default download. Transformer and vocoder weights are dynamic-range INT8; each size bucket is a signature of one model file, sharing weights.
 - `coreai/` (155 MB): Core AI bundle for apps on iOS, iPadOS and visionOS 27 (`.aimodel` assets); it also runs on macOS 27, where the runtime uses the Core ML bundle by default. INT8 weights per output channel with FP16 compute; size buckets are functions of one asset sharing weights. The flow also runs accurately on the Neural Engine (`controls.coreai.neural_engine_assets`), which is the placement for apps that speak in the background; the GPU is the default.
 - `coreml/` (142 MB): Core ML bundle for Apple silicon with iOS 18 / visionOS 2 / macOS 15 and later (compiled `.mlmodelc` assets), with INT8 weights and FP16 compute; GPU by default, CPU in background apps.
-- `onnx/` (262 MB): ONNX Runtime bundle with dynamic INT8 transformer weights; the vocoder stays in full precision.
 - `pytorch/`: duration, vector, adapter/vocoder and G2P weights (no training state), with the release configuration assets.
 - `SHA256SUMS.json`: published-file hashes.
@@ -134,8 +135,8 @@ Every runtime bundle includes the G2P model in a matching precision. The Apple b
 Release validation separates graph accuracy, frontend/timing behaviour and perceived speech quality:
 - Component comparisons against PyTorch during export, and end-to-end comparison of full passages against the PyTorch reference path.
-- The distilled flow at 4 Euler steps compared with the base flow at 8 Heun steps on the same passages, seeds and noise: equal distance to the 32-step output, confirmed by listening across all ten voices.
-- Quantized bundles compared with full precision on real sentences: identical phonemes, phone durations and generated latents within small tolerances, followed by listening across all ten voices.
 - G2P conversions compared with the full-precision G2P on multilingual text.
 - Core ML and Core AI bundles compared with the PyTorch model on every graph and on real passages, on each compute unit they are used with (GPU, CPU, and for Core AI the Neural Engine): identical phonemes apart from two near-tie stress marks also seen in FP16, fewer changed phone durations than the LiteRT bundle, and smaller latent and spectral differences. The Apple exports rewrite three computations into equivalent forms that FP16 and the Neural Engine evaluate exactly (span pooling, the modifier bit decoding, the duration model's combined phone/segment index), and scale the flow's residual stream; in FP32 the rewritten graphs are bit-identical to the trained ones.

 # Scylla's Band v2
+Release **`v2-20261020`**. The [Scylla's Band runtime](https://github.com/lowkeytea/scyllasband) supports this release. Tag `v2-20261020-apple` adds Core ML and Core AI bundles of the same model; the runtime downloads from that tag.
 The model exposes **energy, tension, valence and assertiveness** on **0–4 scales with neutral at 2**. Combining the controls changes the generated delivery; a request is not a guarantee of a particular perceived emotion or exact external scorer value.
 | Angry | `energy=2.3,tension=2.4,valence=1.4,assertiveness=2.4` |
 | Sad | `energy=1.7,tension=1.5,valence=1.4,assertiveness=2` |
+These are moderate styles; how strongly each comes through varies with voice, language and text. The [voice gallery](https://lowkeytea.github.io/scyllasband/#delivery) plays each of them for all fourteen voices.
 `--delivery auto` omits all coordinates. A per-axis `auto` omits just that coordinate; other unspecified fields in a partial request use neutral defaults. Omission is represented by a separate mask, not by the value 2. It is not an automatic text-to-performance planner.
 | Audio | Mono 24 kHz; 100 mel features; 24 latent channels |
 | Mel / latent hop | 256 / 512 samples |
 | Duration model | Hidden size 320; 8 layers; reads the neighbouring text as context |
+| Vector model | Hidden size 640; 16 layers; adaptive layer normalization and QK normalization |
 | Maximum graph phones per sentence | 512 |
 | Trained target length | 64–420 latent frames (about 1.4–9 s) per pass |
 | Latent buckets | 128, 256, 384, 512, 768, 1024 frames |
 ## Voices and language coverage
+Fourteen voices: Ariadne, Felix, Gwen, Ink, Kincade, Max, Orpheus, Poppy, Priya, Rex, Scylla, Stone, Thistle, Tuesday.
+All fourteen have Spanish (`es`), Italian (`it`), French (`fr`), German (`de`) and Vietnamese (`vi`). Ink, Orpheus and Tuesday use British English (`en_gb`); the other eleven use American English (`en_us`). This is **84 trained voice/locale pairs**, not every combination of fourteen voices and seven locale IDs.
 ## Artifacts
+- `litert/` (228 MB): LiteRT bundle; default download. Transformer and vocoder weights are dynamic-range INT8; each size bucket is a signature of one model file, sharing weights.
 - `coreai/` (155 MB): Core AI bundle for apps on iOS, iPadOS and visionOS 27 (`.aimodel` assets); it also runs on macOS 27, where the runtime uses the Core ML bundle by default. INT8 weights per output channel with FP16 compute; size buckets are functions of one asset sharing weights. The flow also runs accurately on the Neural Engine (`controls.coreai.neural_engine_assets`), which is the placement for apps that speak in the background; the GPU is the default.
 - `coreml/` (142 MB): Core ML bundle for Apple silicon with iOS 18 / visionOS 2 / macOS 15 and later (compiled `.mlmodelc` assets), with INT8 weights and FP16 compute; GPU by default, CPU in background apps.
+- `onnx/` (323 MB): ONNX Runtime bundle with dynamic INT8 transformer weights; the vocoder stays in full precision.
+- `webgpu/` (325 MB): ONNX Runtime bundle for ONNX Runtime Web's WebGPU backend, used by the browser demo. Transformer weights are INT8 with float compute (`MatMulNBits`), which runs entirely on the GPU; the vocoder stays in full precision.
 - `pytorch/`: duration, vector, adapter/vocoder and G2P weights (no training state), with the release configuration assets.
 - `SHA256SUMS.json`: published-file hashes.
 Release validation separates graph accuracy, frontend/timing behaviour and perceived speech quality:
 - Component comparisons against PyTorch during export, and end-to-end comparison of full passages against the PyTorch reference path.
+- The distilled flow at 4 Euler steps compared with the base flow on the same passages, seeds and noise: closer to the base flow's 32-step output than the base flow's own 8 Heun steps, and compared blind with the 32-step base flow by listening across all fourteen voices.
+- Quantized bundles compared with full precision on real sentences (48 passages, six voices): identical phonemes; identical phone durations for 97.7% (ONNX), 97.9% (LiteRT) and 99.6% (WebGPU) of phones, the rest one frame apart; generated latents within 2% median relative error; followed by listening across all fourteen voices.
 - G2P conversions compared with the full-precision G2P on multilingual text.
 - Core ML and Core AI bundles compared with the PyTorch model on every graph and on real passages, on each compute unit they are used with (GPU, CPU, and for Core AI the Neural Engine): identical phonemes apart from two near-tie stress marks also seen in FP16, fewer changed phone durations than the LiteRT bundle, and smaller latent and spectral differences. The Apple exports rewrite three computations into equivalent forms that FP16 and the Neural Engine evaluate exactly (span pooling, the modifier bit decoding, the duration model's combined phone/segment index), and scale the flow's residual stream; in FP32 the rewritten graphs are bit-identical to the trained ones.

SHA256SUMS.json CHANGED Viewed

@@ -4,244 +4,16 @@
   "bytes": 11357
  },
  "README.md": {
-  "sha256": "059fae18d14420f7cc376c606059157175c948015a9633ea0c057b6154f765dd",
-  "bytes": 12553
  },
  "THIRD_PARTY_NOTICES.md": {
-  "sha256": "3b793b8ee250cc711e4012ee42e9625a8a7e788bf12cab85a1abbd17b17644ca",
-  "bytes": 2818
- },
- "coreai/assets/fixed_references.json": {
-  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
-  "bytes": 1115042
- },
- "coreai/assets/g2p/export_config.json": {
-  "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
-  "bytes": 928
- },
- "coreai/assets/g2p/language_map.json": {
-  "sha256": "b3db2048b492799f8746b43001e0f7411eb5d359e17cabc6bfce4a7bccbfdd99",
-  "bytes": 844
- },
- "coreai/assets/g2p/normalization.json": {
-  "sha256": "f9ca4fe4e4eca5e02b6f2874f5da6ede8b44b25c71c3c6ac00f96d7397024de6",
-  "bytes": 1090
- },
- "coreai/assets/g2p/phoneme_dict.json": {
-  "sha256": "9cfb40b2fd31d7a32351c33f90230062554a4dc0a44e89f722de211d42ed1875",
-  "bytes": 93
- },
- "coreai/assets/g2p/pronunciation_overrides.json": {
-  "sha256": "a1f311379b76cb44719a2242a93a4046ff5f344cdac613374288234c2817eb59",
-  "bytes": 80
- },
- "coreai/assets/g2p/tokenizer.json": {
-  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
-  "bytes": 7769
- },
- "coreai/assets/languages.json": {
-  "sha256": "f2af2233d0a1c0dbb6e6236fd57592d12da84262fdd1a336ead1ee83a657e2ae",
-  "bytes": 289
- },
- "coreai/assets/phone_vocab.json": {
-  "sha256": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
-  "bytes": 1793
- },
- "coreai/assets/voices.json": {
-  "sha256": "db9aa57e3d5f9f34fed283fb8e79f4c38e218647d33bc4a5c71ea3e8338c1a56",
-  "bytes": 433
- },
- "coreai/coreai/duration_predictor.aimodel/main.hash": {
-  "sha256": "8a93090b36607842c0321f257c7f823cfaee1e27d21f1169647b97fe60614689",
-  "bytes": 32
- },
- "coreai/coreai/duration_predictor.aimodel/main.mlirb": {
-  "sha256": "ac195ab3d8dd35789763adaa1579c7ba38f13a53e7f6ce158493c6dd9013e3c9",
-  "bytes": 10884922
- },
- "coreai/coreai/duration_predictor.aimodel/metadata.json": {
-  "sha256": "8cbd440bab452bfeab9f94ec2ca20fcacd6fa4fba9381c00580e4304f5e98eb6",
-  "bytes": 105
- },
- "coreai/coreai/g2p.aimodel/main.hash": {
-  "sha256": "8bb3d092251c11201ed950315bb0a4349e49c1e9cd6953f57f8a59fc8db148f0",
-  "bytes": 32
- },
- "coreai/coreai/g2p.aimodel/main.mlirb": {
-  "sha256": "0322efaacdef000751cd7cde1cebbaa9b731a85be63c4e5a841785cda7262606",
-  "bytes": 28377277
- },
- "coreai/coreai/g2p.aimodel/metadata.json": {
-  "sha256": "8803aedc1c77719a8f28c087dd45c5893f7756fc80baf2b1709dfa9f55be8a72",
-  "bytes": 105
- },
- "coreai/coreai/vector_context_encoder.aimodel/main.hash": {
-  "sha256": "db6935fcf5d891e6a05b1d888df8f38405341f79756ab7f72483c6ea783e8f70",
-  "bytes": 32
- },
- "coreai/coreai/vector_context_encoder.aimodel/main.mlirb": {
-  "sha256": "0538b39624f01e17f80b5c1640195b7d0082dc82f2b0e294ad5540501d72d8c9",
-  "bytes": 8029273
- },
- "coreai/coreai/vector_context_encoder.aimodel/metadata.json": {
-  "sha256": "40d6747f0a9572dfedf07583a35372a37b8c7ee32b0a64afce43d0ade074e340",
-  "bytes": 105
- },
- "coreai/coreai/vector_estimator.aimodel/main.hash": {
-  "sha256": "38e0d7d0a992bce397e04c03aabef88737478e71b81b02805c5b8c0aa3a35130",
-  "bytes": 32
- },
- "coreai/coreai/vector_estimator.aimodel/main.mlirb": {
-  "sha256": "b45b3c5622c8ebc4dd5a196bd3049291ffb55fbd2735f00f56f3687efc654be9",
-  "bytes": 63149070
- },
- "coreai/coreai/vector_estimator.aimodel/metadata.json": {
-  "sha256": "e84d3e04445c31d3f78b07b3149ba0034e6fb7a00bd0d3bca433d1e359464853",
-  "bytes": 105
- },
- "coreai/coreai/vocoder.aimodel/main.hash": {
-  "sha256": "0146b6f59f58906dee4f23d757cc715cd8daab1588df20dbac89ea7fb48c6ad5",
-  "bytes": 32
- },
- "coreai/coreai/vocoder.aimodel/main.mlirb": {
-  "sha256": "ebd4ca3710eeeaf5829ceb02d35442fc684f4904d0cabf34a833682b599a309c",
-  "bytes": 43811015
- },
- "coreai/coreai/vocoder.aimodel/metadata.json": {
-  "sha256": "c46f8e99ab118a74145ae0c71e72494c79198a1b988458ca6b539e2891577223",
-  "bytes": 105
- },
- "coreai/manifest.json": {
-  "sha256": "d612b45fffe5b3aaaec4edeab12073c8e99e9bd58242d3bf4995659c5eba47b6",
-  "bytes": 21639
- },
- "coreml/assets/fixed_references.json": {
-  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
-  "bytes": 1115042
- },
- "coreml/assets/g2p/export_config.json": {
-  "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
-  "bytes": 928
- },
- "coreml/assets/g2p/language_map.json": {
-  "sha256": "b3db2048b492799f8746b43001e0f7411eb5d359e17cabc6bfce4a7bccbfdd99",
-  "bytes": 844
- },
- "coreml/assets/g2p/normalization.json": {
-  "sha256": "f9ca4fe4e4eca5e02b6f2874f5da6ede8b44b25c71c3c6ac00f96d7397024de6",
-  "bytes": 1090
- },
- "coreml/assets/g2p/phoneme_dict.json": {
-  "sha256": "9cfb40b2fd31d7a32351c33f90230062554a4dc0a44e89f722de211d42ed1875",
-  "bytes": 93
- },
- "coreml/assets/g2p/pronunciation_overrides.json": {
-  "sha256": "a1f311379b76cb44719a2242a93a4046ff5f344cdac613374288234c2817eb59",
-  "bytes": 80
- },
- "coreml/assets/g2p/tokenizer.json": {
-  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
-  "bytes": 7769
- },
- "coreml/assets/languages.json": {
-  "sha256": "f2af2233d0a1c0dbb6e6236fd57592d12da84262fdd1a336ead1ee83a657e2ae",
-  "bytes": 289
- },
- "coreml/assets/phone_vocab.json": {
-  "sha256": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
-  "bytes": 1793
- },
- "coreml/assets/voices.json": {
-  "sha256": "db9aa57e3d5f9f34fed283fb8e79f4c38e218647d33bc4a5c71ea3e8338c1a56",
-  "bytes": 433
- },
- "coreml/coreml/duration_predictor.mlmodelc/analytics/coremldata.bin": {
-  "sha256": "4482c9bb60b1d8a0c2b132ba3dcda5a114b26eeff946e7ae38bd96d66f1e7f41",
-  "bytes": 243
- },
- "coreml/coreml/duration_predictor.mlmodelc/coremldata.bin": {
-  "sha256": "11e6783121ec6660c51ddd5da386e18220f0d43ef20c485459a822adc5d29402",
-  "bytes": 554
- },
- "coreml/coreml/duration_predictor.mlmodelc/model.mil": {
-  "sha256": "b342ee3bcc2acb6b8bd0de42bebfc6c3d83b40e5512156fb26f909fa9ae3d608",
-  "bytes": 138985
- },
- "coreml/coreml/duration_predictor.mlmodelc/weights/weight.bin": {
-  "sha256": "5383e19195d6807cf6632d2c82b51b86aaa947ba51d8042434f73b5039c70ea4",
-  "bytes": 10562880
- },
- "coreml/coreml/g2p.mlmodelc/analytics/coremldata.bin": {
-  "sha256": "70e1596126ef45b7692226041b4f56570d9b381d9f0ad5714c8dee3c745a762e",
-  "bytes": 243
- },
- "coreml/coreml/g2p.mlmodelc/coremldata.bin": {
-  "sha256": "f5fc8bee0186f33fa14d35a59a58ec3393676d07bdf03802a3836d806662cd03",
-  "bytes": 377
- },
- "coreml/coreml/g2p.mlmodelc/model.mil": {
-  "sha256": "d464334a231d99b9a3764d50c28657d1720e25eaa45e39fa4daa401bb89452a1",
-  "bytes": 388409
- },
- "coreml/coreml/g2p.mlmodelc/weights/weight.bin": {
-  "sha256": "045da2991ed2a42ba027aaa6b83813054f410b6eedfdce82f2ab287ec4ca74e2",
-  "bytes": 27346752
- },
- "coreml/coreml/vector_context_encoder.mlmodelc/analytics/coremldata.bin": {
-  "sha256": "d23387f42991cc2675cfa0ed4d8830615bb3f8e760916cce4ba7d3e54c48de21",
-  "bytes": 243
- },
- "coreml/coreml/vector_context_encoder.mlmodelc/coremldata.bin": {
-  "sha256": "118d56856570862aa0e3dde8d3cacdc600d4147da424ec88a22fac56c20cf493",
-  "bytes": 476
- },
- "coreml/coreml/vector_context_encoder.mlmodelc/model.mil": {
-  "sha256": "c3c0df163512f0ed8ab4dbb927054f2cb627980c734419bd96e341540660d372",
-  "bytes": 44809
- },
- "coreml/coreml/vector_context_encoder.mlmodelc/weights/weight.bin": {
-  "sha256": "70da5deb1236e2f5d3250cef65ee0619f1089ebb7c5fc39744d205d45bf24c66",
-  "bytes": 7572352
- },
- "coreml/coreml/vector_estimator.mlmodelc/analytics/coremldata.bin": {
-  "sha256": "13eab20d76b95b33201ab96dd522599eabd282b36db8c231832834862d4385b7",
-  "bytes": 243
- },
- "coreml/coreml/vector_estimator.mlmodelc/coremldata.bin": {
-  "sha256": "d22410a5a41fc0aae79224588c2da505cd4abae60526d06fd989749001741762",
-  "bytes": 3647
- },
- "coreml/coreml/vector_estimator.mlmodelc/model.mil": {
-  "sha256": "0c02ef0f5ff1bc4cfcb712f8b39a55d1f1f25b86aed0f39157bcad8aef0a9fd4",
-  "bytes": 1469004
- },
- "coreml/coreml/vector_estimator.mlmodelc/weights/weight.bin": {
-  "sha256": "dfcd7719b9406f6a181f17f48125c33213196f9e722844f88ecbae5a580a8015",
-  "bytes": 60754048
- },
- "coreml/coreml/vocoder.mlmodelc/analytics/coremldata.bin": {
-  "sha256": "1265ad8d6e6ff623afa255fc3eab94c840e6b372ec6d2ebcacfe9ee053c9aade",
-  "bytes": 243
- },
- "coreml/coreml/vocoder.mlmodelc/coremldata.bin": {
-  "sha256": "04bf9f9de7edfc6ce58e5928a0a1d6669d09095ebc03cb5ca24dd85a416653df",
-  "bytes": 1190
- },
- "coreml/coreml/vocoder.mlmodelc/model.mil": {
-  "sha256": "e7b28b62c2ce75b56bb284333542585cf4abe4ebda50ce72322cbea88dcea53c",
-  "bytes": 969259
- },
- "coreml/coreml/vocoder.mlmodelc/weights/weight.bin": {
-  "sha256": "8d1aa1e85be960d5ad32f751b602c38601bc9517a951be5bd42c991449d3bccc",
-  "bytes": 31043584
- },
- "coreml/manifest.json": {
-  "sha256": "6d9908c3a95d2aff67c2e2ee208e72e94596146a5ac92755a83d470b53c4620c",
-  "bytes": 21531
  },
  "litert/assets/fixed_references.json": {
-  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
-  "bytes": 1115042
  },
  "litert/assets/g2p/export_config.json": {
   "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
@@ -276,36 +48,36 @@
   "bytes": 1793
  },
  "litert/assets/voices.json": {
-  "sha256": "db9aa57e3d5f9f34fed283fb8e79f4c38e218647d33bc4a5c71ea3e8338c1a56",
-  "bytes": 433
  },
  "litert/litert/duration_predictor.tflite": {
-  "sha256": "a7b063e6121ca62bc7e4940822d1016c47999c1688bd7482b27cd7396d5b5bab",
-  "bytes": 11599200
  },
  "litert/litert/g2p.tflite": {
   "sha256": "a416666d1088ccb6fe59b91c7c1c343aad4374f08d4c8e21726fc7aecb79bb0c",
   "bytes": 28626144
  },
  "litert/litert/vector_context_encoder.tflite": {
-  "sha256": "a7d653507b92d8a5102f4281361e8ee8f03cf001b6889e1434ad1e760fbcf057",
-  "bytes": 7763584
  },
  "litert/litert/vector_estimator.tflite": {
-  "sha256": "51f6ce100d8565fa7979718ca34ad0b19d1e14f6525c41c46112dfbd1d5b30cd",
-  "bytes": 76550528
  },
  "litert/litert/vocoder.tflite": {
-  "sha256": "a0410221c7742098fa5efea7f6f7626e522d3a761d84d8bf2e811c9d0de534f3",
-  "bytes": 38613392
  },
  "litert/manifest.json": {
-  "sha256": "7462b9d6ee715d8c44400994d4065b8eb635b724b6bc29ed8cbdf668e662a444",
-  "bytes": 20172
  },
  "onnx/assets/fixed_references.json": {
-  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
-  "bytes": 1115042
  },
  "onnx/assets/g2p/export_config.json": {
   "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
@@ -340,76 +112,76 @@
   "bytes": 1793
  },
  "onnx/assets/voices.json": {
-  "sha256": "db9aa57e3d5f9f34fed283fb8e79f4c38e218647d33bc4a5c71ea3e8338c1a56",
-  "bytes": 433
  },
  "onnx/manifest.json": {
-  "sha256": "544760ec2d22c47666d687f5f162d531a038222ac33fb8aed81042ea8ad06703",
-  "bytes": 20550
  },
  "onnx/onnx/components/duration_predictor.onnx": {
-  "sha256": "7f30552820354c976ba585691ab3c9d94f272df15125b36ef695343801bf6259",
-  "bytes": 285227
  },
  "onnx/onnx/components/shared_weights.bin": {
-  "sha256": "e39b8dbc01078e6d7629057ec3a092fe17a40a090fb0d2839fbbd7dbaee7fbed",
-  "bytes": 208219968
  },
  "onnx/onnx/components/shared_weights.json": {
-  "sha256": "3acfa38030535361101703f2e124ee62061d89ff11c41af1055eeea4cca67765",
-  "bytes": 990204
  },
  "onnx/onnx/components/vector_context_encoder.onnx": {
-  "sha256": "79b04221744e831c80516cfee5120b024b43fba573b9ab943fe2bd89376a0dfd",
-  "bytes": 66364
  },
  "onnx/onnx/components/vector_estimator.onnx": {
-  "sha256": "5dd555761629826c32fea6aa5e3b25aae937cbe827336b8545efb3cd39d98811",
-  "bytes": 2800846
  },
  "onnx/onnx/components/vector_estimator_128.onnx": {
-  "sha256": "ae45282a20cadac7663d30d2e754a60b9164a15e5fbb390a6ec21f9ab4e3d33a",
-  "bytes": 915657
  },
  "onnx/onnx/components/vector_estimator_256.onnx": {
-  "sha256": "007f094ad391f1d444131f84fd09aaf31c9e3a2ab2d3a70af157f9cdc9fd0c95",
-  "bytes": 1184969
  },
  "onnx/onnx/components/vector_estimator_384.onnx": {
-  "sha256": "dcd1337c9024c6bdcba4698f4c93a13d51fed56c5b55fced6fcefe2f6196de4a",
-  "bytes": 1454281
  },
  "onnx/onnx/components/vector_estimator_512.onnx": {
-  "sha256": "8d4c522cfc7d2167dbc7ec4017e14757878b52aedebf7fd522e7a0cfe0a19096",
-  "bytes": 1723593
  },
  "onnx/onnx/components/vector_estimator_768.onnx": {
-  "sha256": "dc186567046dc1eb017329f38f11419efc31f6749b1344fca8abb55c49b6b516",
-  "bytes": 2262218
  },
  "onnx/onnx/components/vocoder.onnx": {
-  "sha256": "9f2343d201977fec1ca729de290ae969ff3457d6fa14bc3daa91dbee6d0088db",
-  "bytes": 122563
  },
  "onnx/onnx/components/vocoder_128.onnx": {
-  "sha256": "44b23f50ed8cd49cbff3a43e435fc35780e3cb63f5169ff5f9613b31869c15e4",
-  "bytes": 122563
  },
  "onnx/onnx/components/vocoder_256.onnx": {
-  "sha256": "6ad741f17902a63aa17d474ad6862cdcfeba796a19f40a892f008cf4f623d884",
-  "bytes": 122562
  },
  "onnx/onnx/components/vocoder_384.onnx": {
-  "sha256": "2fc1a749fd557d17ca352a269dacf042f4ca1da750ef74bc5e4982d2a8f9f533",
-  "bytes": 122563
  },
  "onnx/onnx/components/vocoder_512.onnx": {
-  "sha256": "b7cc541079751236d3612f457c25c45748d986d715e836293cffa23336f31b80",
-  "bytes": 122563
  },
  "onnx/onnx/components/vocoder_768.onnx": {
-  "sha256": "4725a2e9b599a5dd393e5454413282974856224e7913aa3701b6629134bb544c",
-  "bytes": 122564
  },
  "onnx/onnx/g2p/model.onnx": {
   "sha256": "b733b7cedef6ec1c1822771f8b5d7ce430fdc7b4c4d7aac0617044419d0e3296",
@@ -424,19 +196,19 @@
   "bytes": 7769
  },
  "pytorch/condition_index.json": {
-  "sha256": "995750fb8ff573f087f75df851fd38dd7d574e3a47773e8f150bb3b928cf251f",
-  "bytes": 612
  },
  "pytorch/duration.pt": {
-  "sha256": "765e8970c5c014be982a0645129682b47338656ec407373683da78318a77a5dd",
-  "bytes": 43039620
  },
  "pytorch/fixed_references.json": {
-  "sha256": "0120e1cf1ef9b3ee19ed3d269c4dc6f6c47e694a59bcce5bba89abe101cd20e4",
-  "bytes": 786473
  },
  "pytorch/g2p.pt": {
-  "sha256": "c1aff82582456abc7667cf471a045ebaea3296a6b74240cc5e40b4686f643950",
   "bytes": 119207485
  },
  "pytorch/phone_vocab.json": {
@@ -444,15 +216,135 @@
   "bytes": 1793
  },
  "pytorch/release_config.json": {
-  "sha256": "b8e900aad0ec09e4b2e96f0d93542df14e271ea7e57c603c542acdef0f80cc6c",
   "bytes": 842
  },
  "pytorch/vector_estimator.pt": {
-  "sha256": "556caf9f64a97329679ca99b3e6b59e1ecfb8ae312ad2321d10e7ba1e979b1da",
-  "bytes": 267318145
  },
  "pytorch/vocoder.pt": {
-  "sha256": "e31e681d8c524f149efc2a0ea55474c28c1e1c44a5188378ba9d14918db2a15d",
-  "bytes": 106799843
  }
 }

   "bytes": 11357
  },
  "README.md": {
+  "sha256": "a510621d1dd98a0c4d0aa0ee0665c82d7ffce200bea3c566d130d3548ef7f3d4",
+  "bytes": 13095
  },
  "THIRD_PARTY_NOTICES.md": {
+  "sha256": "757a3eca5f8284352d471887af2c31b65c5d23d2a642f0ae77e6a11c150dba39",
+  "bytes": 2829
  },
  "litert/assets/fixed_references.json": {
+  "sha256": "f227c55cab5f6b8cffa33c3dd743e69b26c7e446be5b9cd9a286095cabaa249e",
+  "bytes": 1555673
  },
  "litert/assets/g2p/export_config.json": {
   "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
   "bytes": 1793
  },
  "litert/assets/voices.json": {
+  "sha256": "64b1b317b859e8b98d2802ecc2ff5be7edfcb82d4c641a4aa4a5a4d18fc4841b",
+  "bytes": 613
  },
  "litert/litert/duration_predictor.tflite": {
+  "sha256": "02d815abcae4b3915f7569a1796364a5b7d70adc8e928fc2b7beb33566f6b444",
+  "bytes": 11668736
  },
  "litert/litert/g2p.tflite": {
   "sha256": "a416666d1088ccb6fe59b91c7c1c343aad4374f08d4c8e21726fc7aecb79bb0c",
   "bytes": 28626144
  },
  "litert/litert/vector_context_encoder.tflite": {
+  "sha256": "de37ff6548202785b9a7c3edac609b9d94d11f437cc959bcfff61fc909272e5e",
+  "bytes": 11894656
  },
  "litert/litert/vector_estimator.tflite": {
+  "sha256": "cebf54a5b5009c4bf47982cf370a61b36bf3279ea8c16f7c0d974a4e5534ed56",
+  "bytes": 146540928
  },
  "litert/litert/vocoder.tflite": {
+  "sha256": "bc04828ad6575713d597585b117081c2f0af6185b6ea9eee6c979dd6b94cbf74",
+  "bytes": 38615216
  },
  "litert/manifest.json": {
+  "sha256": "e1c4791b0c9b24e5d3b97763c72b95922617402278e5c959736b3aa8f33c703c",
+  "bytes": 21826
  },
  "onnx/assets/fixed_references.json": {
+  "sha256": "f227c55cab5f6b8cffa33c3dd743e69b26c7e446be5b9cd9a286095cabaa249e",
+  "bytes": 1555673
  },
  "onnx/assets/g2p/export_config.json": {
   "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
   "bytes": 1793
  },
  "onnx/assets/voices.json": {
+  "sha256": "64b1b317b859e8b98d2802ecc2ff5be7edfcb82d4c641a4aa4a5a4d18fc4841b",
+  "bytes": 613
  },
  "onnx/manifest.json": {
+  "sha256": "c29e1429becb22926ce54a0f78c4d704934e9a5d2d6e7bfdc67c6c7023d3c749",
+  "bytes": 22202
  },
  "onnx/onnx/components/duration_predictor.onnx": {
+  "sha256": "5898cff3500e2bb70983d64fb9169d3978640a132eeec9eb9c5ba339c56bb1ed",
+  "bytes": 285360
  },
  "onnx/onnx/components/shared_weights.bin": {
+  "sha256": "a74c6385cadd35d3fb18bbd37b13fc3e4a7a1ffd8b9363455fe5e53fd01a95f3",
+  "bytes": 280418752
  },
  "onnx/onnx/components/shared_weights.json": {
+  "sha256": "704a7eec8b2692f8cffb466ad91adbe02b6bdea7740ec68a717c41953370dd42",
+  "bytes": 1147875
  },
  "onnx/onnx/components/vector_context_encoder.onnx": {
+  "sha256": "0d1b63a2a7fb16960aab4312ce2440bdd6e1cb5d6c848959f5c1ac7f7c0322bd",
+  "bytes": 66883
  },
  "onnx/onnx/components/vector_estimator.onnx": {
+  "sha256": "504e09be116ee26d7923064d7f99373f072c280bbb5168e97eccfb9fda4e280a",
+  "bytes": 3583332
  },
  "onnx/onnx/components/vector_estimator_128.onnx": {
+  "sha256": "2c2df042033c9c7b11645d14d4ddd40bfae1d9850a70d4e8a8df2c3ccea954c3",
+  "bytes": 1225055
  },
  "onnx/onnx/components/vector_estimator_256.onnx": {
+  "sha256": "4e041d4ea61f8d6d27b8f316b988b31ac57743b06ed85ee0e40be08c0857bf29",
+  "bytes": 1561951
  },
  "onnx/onnx/components/vector_estimator_384.onnx": {
+  "sha256": "2ffac242daff773f375b20285aaa03d0c59bbaa86bfb83fe620ad07d5f873281",
+  "bytes": 1898847
  },
  "onnx/onnx/components/vector_estimator_512.onnx": {
+  "sha256": "20f3735423718f6a56ae76b56cfd192c1a9e568cfe9fe181039bcb5b16313880",
+  "bytes": 2235744
  },
  "onnx/onnx/components/vector_estimator_768.onnx": {
+  "sha256": "a14f7e2fa49562ed4c97280121a0b0c9b3ec7f0d795bab56f5ef8efdb7f8e5e0",
+  "bytes": 2909536
  },
  "onnx/onnx/components/vocoder.onnx": {
+  "sha256": "52b5ad8f465931c031392bfb698cd3be579aee094ef564d2c577d89dd1aa06bd",
+  "bytes": 122581
  },
  "onnx/onnx/components/vocoder_128.onnx": {
+  "sha256": "b260cff8e5e69b753c68dc7fef6b7b2894943e0284e80c7b30fc4cf76ea33e72",
+  "bytes": 122580
  },
  "onnx/onnx/components/vocoder_256.onnx": {
+  "sha256": "c6038171611a4e2f8422779b52a63ef5e5d717ea6a40dd6e00b4d6826f730e02",
+  "bytes": 122579
  },
  "onnx/onnx/components/vocoder_384.onnx": {
+  "sha256": "f2eaff8faca9ed11e8a579779e7751df0f00bfaf92913ce23e62690958d0e32b",
+  "bytes": 122580
  },
  "onnx/onnx/components/vocoder_512.onnx": {
+  "sha256": "ebcb8a42f5e0c7344900dfca534401bfb548ed24b4037a08a1ca8c3297810cd0",
+  "bytes": 122580
  },
  "onnx/onnx/components/vocoder_768.onnx": {
+  "sha256": "fbdca0192ce47ea405fec338dc7d499749008789a04b74d4d0f69fc80530db31",
+  "bytes": 122581
  },
  "onnx/onnx/g2p/model.onnx": {
   "sha256": "b733b7cedef6ec1c1822771f8b5d7ce430fdc7b4c4d7aac0617044419d0e3296",
   "bytes": 7769
  },
  "pytorch/condition_index.json": {
+  "sha256": "aee7cca3cb35fae630972663cf7c46e9345612c920ef6168187407110173ddec",
+  "bytes": 776
  },
  "pytorch/duration.pt": {
+  "sha256": "6ca5a04e18924144b9658f0f1d9da271ba1974cab4bd2cf0362b08ed39ef6033",
+  "bytes": 43044740
  },
  "pytorch/fixed_references.json": {
+  "sha256": "523eb475976c4749a8b6bab1ecce982f147bc2b576b3b35544df8173fa35c57a",
+  "bytes": 1095680
  },
  "pytorch/g2p.pt": {
+  "sha256": "008143e12935199d9250b48625a7af3c8ec75acb345a596eeb0c9dde932a46be",
   "bytes": 119207485
  },
  "pytorch/phone_vocab.json": {
   "bytes": 1793
  },
  "pytorch/release_config.json": {
+  "sha256": "03fd9da0df1bb3e6e9c57b851aeb7170b93435791b74b46652a952136efaa024",
   "bytes": 842
  },
  "pytorch/vector_estimator.pt": {
+  "sha256": "08f619ee9393b1d179636e805dd738ed09025ef01426acf810de19529a83c1cb",
+  "bytes": 533866961
  },
  "pytorch/vocoder.pt": {
+  "sha256": "4bc62614f579e245a00d06ae9cc259ca177c35b8713b042af7bb09709f6ba9cb",
+  "bytes": 106805987
+ },
+ "webgpu/assets/fixed_references.json": {
+  "sha256": "f227c55cab5f6b8cffa33c3dd743e69b26c7e446be5b9cd9a286095cabaa249e",
+  "bytes": 1555673
+ },
+ "webgpu/assets/g2p/export_config.json": {
+  "sha256": "0ec5ea81f346521f3a5c9c4a90b939c6b5143b1a11cc5e74b9902969db56e568",
+  "bytes": 928
+ },
+ "webgpu/assets/g2p/language_map.json": {
+  "sha256": "b3db2048b492799f8746b43001e0f7411eb5d359e17cabc6bfce4a7bccbfdd99",
+  "bytes": 844
+ },
+ "webgpu/assets/g2p/normalization.json": {
+  "sha256": "f9ca4fe4e4eca5e02b6f2874f5da6ede8b44b25c71c3c6ac00f96d7397024de6",
+  "bytes": 1090
+ },
+ "webgpu/assets/g2p/phoneme_dict.json": {
+  "sha256": "9cfb40b2fd31d7a32351c33f90230062554a4dc0a44e89f722de211d42ed1875",
+  "bytes": 93
+ },
+ "webgpu/assets/g2p/pronunciation_overrides.json": {
+  "sha256": "a1f311379b76cb44719a2242a93a4046ff5f344cdac613374288234c2817eb59",
+  "bytes": 80
+ },
+ "webgpu/assets/g2p/tokenizer.json": {
+  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
+  "bytes": 7769
+ },
+ "webgpu/assets/languages.json": {
+  "sha256": "f2af2233d0a1c0dbb6e6236fd57592d12da84262fdd1a336ead1ee83a657e2ae",
+  "bytes": 289
+ },
+ "webgpu/assets/phone_vocab.json": {
+  "sha256": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
+  "bytes": 1793
+ },
+ "webgpu/assets/voices.json": {
+  "sha256": "64b1b317b859e8b98d2802ecc2ff5be7edfcb82d4c641a4aa4a5a4d18fc4841b",
+  "bytes": 613
+ },
+ "webgpu/manifest.json": {
+  "sha256": "4e7666bd543b8bdfb28b41c74da317c26de9f08dac0fc11f36c9a36c1600f204",
+  "bytes": 22506
+ },
+ "webgpu/onnx/components/duration_predictor.onnx": {
+  "sha256": "13a3cf33aeccf04cd2ad749ddd1eef112bc4f28d8e4a9ce94353c9e61a53f542",
+  "bytes": 181227
+ },
+ "webgpu/onnx/components/shared_weights.bin": {
+  "sha256": "411e81ca4462cb3586bfa7011cd0e3747e127d7111deb1c76dbc5ac4cb7afd64",
+  "bytes": 284376512
+ },
+ "webgpu/onnx/components/shared_weights.json": {
+  "sha256": "04ef0d699d5694e0f25bb7043e61ec221e74281c8670c1c80e71bdf708e38f75",
+  "bytes": 1085876
+ },
+ "webgpu/onnx/components/vector_context_encoder.onnx": {
+  "sha256": "387cf72a2d5f01f04fc438d4979c619e11b423144825f4cee5ed17217ec100c2",
+  "bytes": 39672
+ },
+ "webgpu/onnx/components/vector_estimator.onnx": {
+  "sha256": "83e550d7b48e7ad263fea5d2bdb594d481f7db50196d3e644ffd4ab6bb9e940c",
+  "bytes": 3318247
+ },
+ "webgpu/onnx/components/vector_estimator_128.onnx": {
+  "sha256": "5fd4d3f7543518c33b5f48e9a5dad4625339ce1ac7d534374b4c87ddf5fd8446",
+  "bytes": 959970
+ },
+ "webgpu/onnx/components/vector_estimator_256.onnx": {
+  "sha256": "e557ba7883c22ddbcdcbad3a32f461ef24085f766c4fff6731018d5376d0b977",
+  "bytes": 1296866
+ },
+ "webgpu/onnx/components/vector_estimator_384.onnx": {
+  "sha256": "f8ab82b9d8c0da64345a86993bb56b23b48caf510d41ac5740245838d9db3c7f",
+  "bytes": 1633762
+ },
+ "webgpu/onnx/components/vector_estimator_512.onnx": {
+  "sha256": "d235d5297d7eb5fbc75facb3e0904e282084af260c19457ab07a173d025eeed7",
+  "bytes": 1970658
+ },
+ "webgpu/onnx/components/vector_estimator_768.onnx": {
+  "sha256": "4d9967b7df1bd0cec9997256b5bc25e02e5ebd84993ac5fe9e67032412540fb0",
+  "bytes": 2644451
+ },
+ "webgpu/onnx/components/vocoder.onnx": {
+  "sha256": "17639fc15ec709018616ee73452c3d5b6809911fdbe033f81a16cdb332fc6284",
+  "bytes": 122578
+ },
+ "webgpu/onnx/components/vocoder_128.onnx": {
+  "sha256": "995f45b8a3fec92b6ad9fa01c949280955542921b7702cf2307b73e967a75872",
+  "bytes": 122577
+ },
+ "webgpu/onnx/components/vocoder_256.onnx": {
+  "sha256": "c69a86241573e86d6a74514d214a5aabfce6523dcedab90011c79aaa791e88d7",
+  "bytes": 122576
+ },
+ "webgpu/onnx/components/vocoder_384.onnx": {
+  "sha256": "3473c5a9f9a43604c9f2cdbcb8b5d2bf4185c4cecbdaf5b5c2fc99ae564fa719",
+  "bytes": 122577
+ },
+ "webgpu/onnx/components/vocoder_512.onnx": {
+  "sha256": "1539bc19030ba6e6d266f5df48bbce4674d3cab3200b5d3f1001d4ee8564b098",
+  "bytes": 122577
+ },
+ "webgpu/onnx/components/vocoder_768.onnx": {
+  "sha256": "0d1a80010d0041f5a87276d0f65a5e1d3499640a66dfeecc8d71a0409d72bada",
+  "bytes": 122578
+ },
+ "webgpu/onnx/g2p/model.onnx": {
+  "sha256": "8090878377b02c22b05d8d9a4e52ac2782642fb026e5d6b2471f2c61fc5eb204",
+  "bytes": 43687646
+ },
+ "webgpu/onnx/g2p/phoneme_dict.json": {
+  "sha256": "9cfb40b2fd31d7a32351c33f90230062554a4dc0a44e89f722de211d42ed1875",
+  "bytes": 93
+ },
+ "webgpu/onnx/g2p/tokenizer.json": {
+  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
+  "bytes": 7769
  }
 }

THIRD_PARTY_NOTICES.md CHANGED Viewed

@@ -5,7 +5,7 @@ store objects defined by, the following MIT-licensed projects.
 ## Vocos
-The 24 kHz waveform decoder in every bundle (`litert/`, `onnx/`, `coreml/`, `coreai/`) and in `pytorch/vocoder.pt` is
 fine-tuned from the Vocos model `charactr/vocos-mel-24khz` (https://github.com/gemelo-ai/vocos).
 ```

 ## Vocos
+The 24 kHz waveform decoder in every bundle (`litert/`, `onnx/`, `webgpu/`, `coreml/`, `coreai/`) and in `pytorch/vocoder.pt` is
 fine-tuned from the Vocos model `charactr/vocos-mel-24khz` (https://github.com/gemelo-ai/vocos).
 ```

coreai/coreai/duration_predictor.aimodel/main.hash DELETED Viewed

	@@ -1 +0,0 @@
1	- �Z��5x�c��yǺ8�:S��ݐ��

coreai/coreai/duration_predictor.aimodel/main.mlirb DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ac195ab3d8dd35789763adaa1579c7ba38f13a53e7f6ce158493c6dd9013e3c9
-size 10884922

coreai/coreai/duration_predictor.aimodel/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "assetVersion" : "2.0",
-  "producer" : "coreai-core 1.0.0b3",
-  "creationDate" : "20261006T145736Z"
-}

coreai/coreai/g2p.aimodel/main.hash DELETED Viewed

Binary file (32 Bytes)

coreai/coreai/g2p.aimodel/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "producer" : "coreai-core 1.0.0b3",
-  "creationDate" : "20261006T145734Z",
-  "assetVersion" : "2.0"
-}

coreai/coreai/vector_context_encoder.aimodel/main.hash DELETED Viewed

Binary file (32 Bytes)

coreai/coreai/vector_context_encoder.aimodel/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "producer" : "coreai-core 1.0.0b3",
-  "assetVersion" : "2.0",
-  "creationDate" : "20261006T145737Z"
-}

coreai/coreai/vector_estimator.aimodel/main.hash DELETED Viewed

	@@ -1 +0,0 @@
1	- �[<V"��Zk��_�'5�V�h~�eK�

coreai/coreai/vector_estimator.aimodel/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "assetVersion" : "2.0",
-  "producer" : "coreai-core 1.0.0b3",
-  "creationDate" : "20261006T145750Z"
-}

coreai/coreai/vocoder.aimodel/main.hash DELETED Viewed

	@@ -1 +0,0 @@
1	- ��7��TB�hOI�ʿ4�3h+Y�0�

coreai/coreai/vocoder.aimodel/main.mlirb DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ebd4ca3710eeeaf5829ceb02d35442fc684f4904d0cabf34a833682b599a309c
-size 43811015

coreai/coreai/vocoder.aimodel/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "assetVersion" : "2.0",
-  "creationDate" : "20261006T145759Z",
-  "producer" : "coreai-core 1.0.0b3"
-}

coreml/assets/fixed_references.json DELETED Viewed

The diff for this file is too large to render. See raw diff

coreml/assets/g2p/export_config.json DELETED Viewed

@@ -1,33 +0,0 @@
-{
-  "batch_dynamic": true,
-  "format": "scyllasband_g2p",
-  "input_granularity": "phrase",
-  "languages": [
-    "de",
-    "en_gb",
-    "en_us",
-    "es",
-    "fr",
-    "it",
-    "vi"
-  ],
-  "model_type": "transformer",
-  "onnx_export_batch_size": 4,
-  "onnx_export_layout": "batch_first",
-  "onnx_export_max_batch_size": 256,
-  "onnx_exporter": "dynamo",
-  "onnx_fp32_model": "model.onnx",
-  "onnx_model": "model.onnx",
-  "onnx_padded_output": "ctc_blank",
-  "onnx_quantization": "none",
-  "onnx_quantized_parity_tolerance": 3.0,
-  "opset": 18,
-  "phoneme_token_mode": "char",
-  "phrase_rows_per_language": 200000,
-  "preserve_commas": true,
-  "punctuation_silence_target": "explicit_silence",
-  "release_role": "fp32_bundle_frontend",
-  "source_checkpoint": "output/deepphonemizer_multilingual/model_v10_7lang/checkpoints/model_step_250k.pt",
-  "training_mix": "phrase_word_chunked_clean",
-  "word_rows_per_language": 200000
-}

coreml/assets/g2p/language_map.json DELETED Viewed

@@ -1,47 +0,0 @@
-{
-  "american": "en_us",
-  "american_english": "en_us",
-  "british": "en_gb",
-  "british_english": "en_gb",
-  "de": "de",
-  "de_de": "de",
-  "deu": "de",
-  "en": "en_us",
-  "en-gb": "en_gb",
-  "en-uk": "en_gb",
-  "en-us": "en_us",
-  "en_gb": "en_gb",
-  "en_uk": "en_gb",
-  "en_us": "en_us",
-  "eng": "en_us",
-  "eng_gb": "en_gb",
-  "eng_us": "en_us",
-  "english": "en_us",
-  "english_gb": "en_gb",
-  "english_uk": "en_gb",
-  "english_us": "en_us",
-  "es": "es",
-  "es_es": "es",
-  "es_mx": "es",
-  "fr": "fr",
-  "fr_fr": "fr",
-  "fra": "fr",
-  "fre": "fr",
-  "french": "fr",
-  "gb": "en_gb",
-  "ger": "de",
-  "german": "de",
-  "it": "it",
-  "it_it": "it",
-  "ita": "it",
-  "italian": "it",
-  "spa": "es",
-  "spanish": "es",
-  "uk": "en_gb",
-  "us": "en_us",
-  "vi": "vi",
-  "vi_hn": "vi",
-  "vi_vn": "vi",
-  "vie": "vi",
-  "vietnamese": "vi"
-}

coreml/assets/g2p/normalization.json DELETED Viewed

@@ -1,40 +0,0 @@
-{
-  "contract": {
-    "colon_semicolon_classes": "preserved",
-    "dot_runs_two_or_more": "ellipsis_three_dots",
-    "homogeneous_terminal_runs": "single_mark",
-    "lexical_hyphen": "space_no_pause",
-    "mixed_question_exclamation": "leading_mark",
-    "non_acoustic_modifiers": [
-      "stress",
-      "length",
-      "unicode_combining"
-    ],
-    "ok_spoken_form": "okay",
-    "punctuation_silence_target": "explicit_silence",
-    "syntactic_dash": "em_dash_pause",
-    "training_phone_source": "espeak",
-    "version": "scyllasband_spoken_text_v2_2026_08_05"
-  },
-  "contract_sha256": "d8fd1b2415d0d100d1c70237a4469117ee6b73d0e4da703e4654c83192779cc4",
-  "contract_version": "scyllasband_spoken_text_v2_2026_08_05",
-  "expand_currency": true,
-  "expand_dates": true,
-  "expand_numbers": true,
-  "expand_ordinals": true,
-  "expand_percentages": true,
-  "expand_times": true,
-  "languages": [
-    "en",
-    "en_us",
-    "en_gb",
-    "es",
-    "it",
-    "fr",
-    "de",
-    "vi"
-  ],
-  "normalize_at_sign": true,
-  "normalize_punctuation": true,
-  "normalizer": "SpokenTextNormalizer"
-}

coreml/assets/g2p/pronunciation_overrides.json DELETED Viewed

@@ -1,4 +0,0 @@
-{
-  "format": "scyllasband_g2p_pronunciation_overrides_v1",
-  "languages": {}
-}

coreml/assets/languages.json DELETED Viewed

@@ -1,30 +0,0 @@
-[
-  {
-    "id": "en_us",
-    "index": 0
-  },
-  {
-    "id": "en_gb",
-    "index": 1
-  },
-  {
-    "id": "es",
-    "index": 2
-  },
-  {
-    "id": "it",
-    "index": 3
-  },
-  {
-    "id": "fr",
-    "index": 4
-  },
-  {
-    "id": "de",
-    "index": 5
-  },
-  {
-    "id": "vi",
-    "index": 6
-  }
-]

coreml/assets/phone_vocab.json DELETED Viewed

@@ -1,111 +0,0 @@
-{
-  "schema_version": 1,
-  "mode": "scyllasband_ipa_phone_symbols",
-  "special_tokens": [
-    "<pad>",
-    "<unk>",
-    "<sil>",
-    "<pause_comma>",
-    "<pause_semicolon>",
-    "<pause_colon>",
-    "<pause_dash>",
-    "<ellipsis>",
-    "<end_stmt>",
-    "<end_question>",
-    "<end_exclaim>",
-    "<ctx_sentence_start>",
-    "<ctx_continuation>",
-    "<ctx_sentence_end>",
-    "<ctx_chunk_continue>"
-  ],
-  "vocab_size": 87,
-  "token_to_id": {
-    "<pad>": 0,
-    "<unk>": 1,
-    "<sil>": 2,
-    "<pause_comma>": 3,
-    "<pause_semicolon>": 4,
-    "<pause_colon>": 5,
-    "<pause_dash>": 6,
-    "<ellipsis>": 7,
-    "<end_stmt>": 8,
-    "<end_question>": 9,
-    "<end_exclaim>": 10,
-    "<ctx_sentence_start>": 11,
-    "<ctx_continuation>": 12,
-    "<ctx_sentence_end>": 13,
-    "<ctx_chunk_continue>": 14,
-    "1": 15,
-    "2": 16,
-    "4": 17,
-    "5": 18,
-    "6": 19,
-    "7": 20,
-    "a": 21,
-    "b": 22,
-    "c": 23,
-    "d": 24,
-    "e": 25,
-    "f": 26,
-    "h": 27,
-    "i": 28,
-    "j": 29,
-    "k": 30,
-    "l": 31,
-    "m": 32,
-    "n": 33,
-    "o": 34,
-    "p": 35,
-    "r": 36,
-    "s": 37,
-    "t": 38,
-    "u": 39,
-    "v": 40,
-    "w": 41,
-    "x": 42,
-    "y": 43,
-    "z": 44,
-    "æ": 45,
-    "ç": 46,
-    "ð": 47,
-    "ø": 48,
-    "ŋ": 49,
-    "œ": 50,
-    "ɐ": 51,
-    "ɑ": 52,
-    "ɒ": 53,
-    "ɔ": 54,
-    "ɗ": 55,
-    "ə": 56,
-    "ɚ": 57,
-    "ɛ": 58,
-    "ɜ": 59,
-    "ɟ": 60,
-    "ɡ": 61,
-    "ɣ": 62,
-    "ɪ": 63,
-    "ɯ": 64,
-    "ɲ": 65,
-    "ɹ": 66,
-    "ɾ": 67,
-    "ʁ": 68,
-    "ʃ": 69,
-    "ʊ": 70,
-    "ʌ": 71,
-    "ʎ": 72,
-    "ʐ": 73,
-    "ʒ": 74,
-    "ʔ": 75,
-    "ʝ": 76,
-    "ʲ": 77,
-    "ˈ": 78,
-    "ˌ": 79,
-    "ː": 80,
-    "̃": 81,
-    "̩": 82,
-    "̪": 83,
-    "β": 84,
-    "θ": 85,
-    "ᵻ": 86
-  }
-}

coreml/assets/voices.json DELETED Viewed

@@ -1,42 +0,0 @@
-[
-  {
-    "id": "ariadne",
-    "index": 0
-  },
-  {
-    "id": "felix",
-    "index": 1
-  },
-  {
-    "id": "gwen",
-    "index": 2
-  },
-  {
-    "id": "ink",
-    "index": 3
-  },
-  {
-    "id": "max",
-    "index": 4
-  },
-  {
-    "id": "orpheus",
-    "index": 5
-  },
-  {
-    "id": "rex",
-    "index": 6
-  },
-  {
-    "id": "scylla",
-    "index": 7
-  },
-  {
-    "id": "stone",
-    "index": 8
-  },
-  {
-    "id": "tuesday",
-    "index": 9
-  }
-]

coreml/coreml/duration_predictor.mlmodelc/model.mil DELETED Viewed

The diff for this file is too large to render. See raw diff

coreml/coreml/duration_predictor.mlmodelc/weights/weight.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:5383e19195d6807cf6632d2c82b51b86aaa947ba51d8042434f73b5039c70ea4
-size 10562880

coreml/coreml/g2p.mlmodelc/model.mil DELETED Viewed

The diff for this file is too large to render. See raw diff

coreml/coreml/g2p.mlmodelc/weights/weight.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:045da2991ed2a42ba027aaa6b83813054f410b6eedfdce82f2ab287ec4ca74e2
-size 27346752

coreml/coreml/vector_context_encoder.mlmodelc/analytics/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d23387f42991cc2675cfa0ed4d8830615bb3f8e760916cce4ba7d3e54c48de21
-size 243

coreml/coreml/vector_context_encoder.mlmodelc/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:118d56856570862aa0e3dde8d3cacdc600d4147da424ec88a22fac56c20cf493
-size 476

coreml/coreml/vector_context_encoder.mlmodelc/model.mil DELETED Viewed

@@ -1,267 +0,0 @@
-program(1.3)
-[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}})]
-{
-    func main<ios18>(tensor<int32, [1, 512]> span_context_mask, tensor<int32, [1, 512]> span_context_phone_ids, tensor<int32, [1, 512]> span_context_segment_ids) {
-            int32 var_15 = const()[name = string("op_15"), val = int32(0)];
-            tensor<bool, [1, 512]> span_context_mask_1 = not_equal(x = span_context_mask, y = var_15)[name = string("span_context_mask")];
-            int32 var_30 = const()[name = string("op_30"), val = int32(-1)];
-            int32 var_35 = const()[name = string("op_35"), val = int32(2)];
-            int32 var_36 = const()[name = string("op_36"), val = int32(86)];
-            int32 var_38 = const()[name = string("op_38"), val = int32(1)];
-            int32 var_40 = const()[name = string("op_40"), val = int32(0)];
-            tensor<int32, [1, 512]> var_55 = maximum(x = span_context_phone_ids, y = var_40)[name = string("op_55")];
-            tensor<int32, [1, 512]> input_1 = minimum(x = var_55, y = var_36)[name = string("input_1")];
-            tensor<int32, [1, 512]> var_59 = maximum(x = span_context_segment_ids, y = var_40)[name = string("op_59")];
-            tensor<int32, [1, 512]> input_3 = minimum(x = var_59, y = var_35)[name = string("input_3")];
-            string var_64_dtype_0 = const()[name = string("op_64_dtype_0"), val = string("int32")];
-            bool var_65_exclusive_0 = const()[name = string("op_65_exclusive_0"), val = bool(false)];
-            bool var_65_reverse_0 = const()[name = string("op_65_reverse_0"), val = bool(false)];
-            tensor<int32, [1, 512]> var_64 = cast(dtype = var_64_dtype_0, x = span_context_mask_1)[name = string("cast_13")];
-            tensor<int32, [1, 512]> var_65 = cumsum(axis = var_38, exclusive = var_65_exclusive_0, reverse = var_65_reverse_0, x = var_64)[name = string("op_65")];
-            int32 var_66 = const()[name = string("op_66"), val = int32(1)];
-            tensor<int32, [1, 512]> var_67 = sub(x = var_65, y = var_66)[name = string("op_67")];
-            tensor<int32, [1, 512]> input_5 = maximum(x = var_67, y = var_40)[name = string("input_5")];
-            int32 var_70_batch_dims_0 = const()[name = string("op_70_batch_dims_0"), val = int32(0)];
-            bool var_70_validate_indices_0 = const()[name = string("op_70_validate_indices_0"), val = bool(false)];
-            tensor<fp16, [87, 512]> module_flow_phone_embed_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [87, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [87, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44672))))[name = string("module_flow_phone_embed_weight_to_fp16_quantized")];
-            string input_1_to_int16_dtype_0 = const()[name = string("input_1_to_int16_dtype_0"), val = string("int16")];
-            string cast_24_dtype_0 = const()[name = string("cast_24_dtype_0"), val = string("int32")];
-            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
-            tensor<int16, [1, 512]> input_1_to_int16 = cast(dtype = input_1_to_int16_dtype_0, x = input_1)[name = string("cast_12")];
-            tensor<int32, [1, 512]> cast_24 = cast(dtype = cast_24_dtype_0, x = input_1_to_int16)[name = string("cast_11")];
-            tensor<bool, [1, 512]> greater_equal_0 = greater_equal(x = cast_24, y = greater_equal_0_y_0)[name = string("greater_equal_0")];
-            int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(87)];
-            tensor<int32, [1, 512]> add_0 = add(x = cast_24, y = slice_by_index_0)[name = string("add_0")];
-            tensor<int32, [1, 512]> select_0 = select(a = cast_24, b = add_0, cond = greater_equal_0)[name = string("select_0")];
-            string select_0_to_int16_dtype_0 = const()[name = string("select_0_to_int16_dtype_0"), val = string("int16")];
-            string cast_0_dtype_0 = const()[name = string("cast_0_dtype_0"), val = string("int32")];
-            int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)];
-            tensor<int16, [1, 512]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = string("cast_10")];
-            tensor<int32, [1, 512]> cast_0 = cast(dtype = cast_0_dtype_0, x = select_0_to_int16)[name = string("cast_9")];
-            tensor<bool, [1, 512]> greater_equal_0_1 = greater_equal(x = cast_0, y = greater_equal_0_y_0_1)[name = string("greater_equal_0_1")];
-            int32 slice_by_index_0_1 = const()[name = string("slice_by_index_0_1"), val = int32(87)];
-            tensor<int32, [1, 512]> add_0_1 = add(x = cast_0, y = slice_by_index_0_1)[name = string("add_0_1")];
-            tensor<int32, [1, 512]> select_0_1 = select(a = cast_0, b = add_0_1, cond = greater_equal_0_1)[name = string("select_0_1")];
-            int32 op_70_cast_fp16_cast_uint16_cast_uint16_axis_0 = const()[name = string("op_70_cast_fp16_cast_uint16_cast_uint16_axis_0"), val = int32(0)];
-            tensor<fp16, [1, 512, 512]> op_70_cast_fp16_cast_uint16_cast_uint16 = gather(axis = op_70_cast_fp16_cast_uint16_cast_uint16_axis_0, batch_dims = var_70_batch_dims_0, indices = select_0_1, validate_indices = var_70_validate_indices_0, x = module_flow_phone_embed_weight_to_fp16_quantized)[name = string("op_70_cast_fp16_cast_uint16_cast_uint16")];
-            int32 var_72_axis_0 = const()[name = string("op_72_axis_0"), val = int32(0)];
-            int32 var_72_batch_dims_0 = const()[name = string("op_72_batch_dims_0"), val = int32(0)];
-            bool var_72_validate_indices_0 = const()[name = string("op_72_validate_indices_0"), val = bool(false)];
-            tensor<fp16, [3, 512]> module_flow_span_segment_embed_weight_to_fp16 = const()[name = string("module_flow_span_segment_embed_weight_to_fp16"), val = tensor<fp16, [3, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44928)))];
-            string input_3_to_uint16_dtype_0 = const()[name = string("input_3_to_uint16_dtype_0"), val = string("uint16")];
-            tensor<uint16, [1, 512]> input_3_to_uint16 = cast(dtype = input_3_to_uint16_dtype_0, x = input_3)[name = string("cast_8")];
-            tensor<fp16, [1, 512, 512]> var_72_cast_fp16_cast_uint16 = gather(axis = var_72_axis_0, batch_dims = var_72_batch_dims_0, indices = input_3_to_uint16, validate_indices = var_72_validate_indices_0, x = module_flow_span_segment_embed_weight_to_fp16)[name = string("op_72_cast_fp16_cast_uint16")];
-            tensor<fp16, [1, 512, 512]> var_73_cast_fp16 = add(x = op_70_cast_fp16_cast_uint16_cast_uint16, y = var_72_cast_fp16_cast_uint16)[name = string("op_73_cast_fp16")];
-            int32 var_75_axis_0 = const()[name = string("op_75_axis_0"), val = int32(0)];
-            int32 var_75_batch_dims_0 = const()[name = string("op_75_batch_dims_0"), val = int32(0)];
-            bool var_75_validate_indices_0 = const()[name = string("op_75_validate_indices_0"), val = bool(false)];
-            tensor<fp16, [768, 512]> module_flow_span_position_embed_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [768, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48064))), scale = tensor<fp16, [768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441344))))[name = string("module_flow_span_position_embed_weight_to_fp16_quantized")];
-            string input_5_to_uint16_dtype_0 = const()[name = string("input_5_to_uint16_dtype_0"), val = string("uint16")];
-            tensor<uint16, [1, 512]> input_5_to_uint16 = cast(dtype = input_5_to_uint16_dtype_0, x = input_5)[name = string("cast_7")];
-            tensor<fp16, [1, 512, 512]> var_75_cast_fp16_cast_uint16 = gather(axis = var_75_axis_0, batch_dims = var_75_batch_dims_0, indices = input_5_to_uint16, validate_indices = var_75_validate_indices_0, x = module_flow_span_position_embed_weight_to_fp16_quantized)[name = string("op_75_cast_fp16_cast_uint16")];
-            tensor<fp16, [1, 512, 512]> hidden_1_cast_fp16 = add(x = var_73_cast_fp16, y = var_75_cast_fp16_cast_uint16)[name = string("hidden_1_cast_fp16")];
-            tensor<int32, [1]> var_82_axes_0 = const()[name = string("op_82_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<bool, [1, 512, 1]> var_82 = expand_dims(axes = var_82_axes_0, x = span_context_mask_1)[name = string("op_82")];
-            tensor<bool, [1, 512, 1]> var_83 = logical_not(x = var_82)[name = string("op_83")];
-            fp16 var_18_to_fp16 = const()[name = string("op_18_to_fp16"), val = fp16(0x0p+0)];
-            tensor<fp16, [1, 512, 512]> hidden_3_cast_fp16 = select(a = var_18_to_fp16, b = hidden_1_cast_fp16, cond = var_83)[name = string("hidden_3_cast_fp16")];
-            tensor<int32, [1]> input_7_axes_0 = const()[name = string("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_norm_attention_weight_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_norm_attention_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(442944)))];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_norm_attention_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_norm_attention_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444032)))];
-            fp16 var_28_to_fp16 = const()[name = string("op_28_to_fp16"), val = fp16(0x1.5p-17)];
-            tensor<fp16, [1, 512, 512]> input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = module_flow_span_ordered_encoder_blocks_0_norm_attention_bias_to_fp16, epsilon = var_28_to_fp16, gamma = module_flow_span_ordered_encoder_blocks_0_norm_attention_weight_to_fp16, x = hidden_3_cast_fp16)[name = string("input_7_cast_fp16")];
-            tensor<fp16, [1536, 512]> module_flow_span_ordered_encoder_blocks_0_qkv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(445120))), scale = tensor<fp16, [1536, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1231616))))[name = string("module_flow_span_ordered_encoder_blocks_0_qkv_weight_to_fp16_quantized")];
-            tensor<fp16, [1536]> module_flow_span_ordered_encoder_blocks_0_qkv_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_qkv_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234752)))];
-            tensor<fp16, [1, 512, 1536]> linear_0_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_0_qkv_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_0_qkv_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("linear_0_cast_fp16")];
-            tensor<int32, [5]> var_100 = const()[name = string("op_100"), val = tensor<int32, [5]>([1, 512, 3, 8, 64])];
-            tensor<fp16, [1, 512, 3, 8, 64]> qkv_1_cast_fp16 = reshape(shape = var_100, x = linear_0_cast_fp16)[name = string("qkv_1_cast_fp16")];
-            tensor<int32, [3]> var_102_split_sizes_0 = const()[name = string("op_102_split_sizes_0"), val = tensor<int32, [3]>([1, 1, 1])];
-            int32 var_102_axis_0 = const()[name = string("op_102_axis_0"), val = int32(2)];
-            tensor<fp16, [1, 512, 1, 8, 64]> var_102_cast_fp16_0, tensor<fp16, [1, 512, 1, 8, 64]> var_102_cast_fp16_1, tensor<fp16, [1, 512, 1, 8, 64]> var_102_cast_fp16_2 = split(axis = var_102_axis_0, split_sizes = var_102_split_sizes_0, x = qkv_1_cast_fp16)[name = string("op_102_cast_fp16")];
-            tensor<int32, [1]> squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = var_102_cast_fp16_0)[name = string("squeeze_0_cast_fp16")];
-            tensor<int32, [1]> squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = var_102_cast_fp16_1)[name = string("squeeze_1_cast_fp16")];
-            tensor<int32, [1]> squeeze_2_axes_0 = const()[name = string("squeeze_2_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_2_cast_fp16 = squeeze(axes = squeeze_2_axes_0, x = var_102_cast_fp16_2)[name = string("squeeze_2_cast_fp16")];
-            tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            bool var_110_transpose_x_0 = const()[name = string("op_110_transpose_x_0"), val = bool(false)];
-            bool var_110_transpose_y_0 = const()[name = string("op_110_transpose_y_0"), val = bool(false)];
-            tensor<int32, [4]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<int32, [4]> transpose_9_perm_0 = const()[name = string("transpose_9_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 64, 512]> transpose_9 = transpose(perm = transpose_9_perm_0, x = squeeze_1_cast_fp16)[name = string("transpose_18")];
-            tensor<fp16, [1, 8, 512, 64]> transpose_8 = transpose(perm = transpose_8_perm_0, x = squeeze_0_cast_fp16)[name = string("transpose_19")];
-            tensor<fp16, [1, 8, 512, 512]> var_110_cast_fp16 = matmul(transpose_x = var_110_transpose_x_0, transpose_y = var_110_transpose_y_0, x = transpose_8, y = transpose_9)[name = string("op_110_cast_fp16")];
-            fp16 var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = fp16(0x1p-3)];
-            tensor<fp16, [1, 8, 512, 512]> scores_1_cast_fp16 = mul(x = var_110_cast_fp16, y = var_111_to_fp16)[name = string("scores_1_cast_fp16")];
-            tensor<int32, [1]> var_114_axes_0 = const()[name = string("op_114_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<bool, [1, 1, 512]> var_114 = expand_dims(axes = var_114_axes_0, x = span_context_mask_1)[name = string("op_114")];
-            tensor<int32, [1]> var_115_axes_0 = const()[name = string("op_115_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<bool, [1, 1, 1, 512]> var_115 = expand_dims(axes = var_115_axes_0, x = var_114)[name = string("op_115")];
-            tensor<bool, [1, 1, 1, 512]> var_117 = logical_not(x = var_115)[name = string("op_117")];
-            fp16 var_23_to_fp16 = const()[name = string("op_23_to_fp16"), val = fp16(-inf)];
-            tensor<fp16, [1, 8, 512, 512]> scores_3_cast_fp16 = select(a = var_23_to_fp16, b = scores_1_cast_fp16, cond = var_117)[name = string("scores_3_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> var_119_cast_fp16 = softmax(axis = var_30, x = scores_3_cast_fp16)[name = string("op_119_cast_fp16")];
-            string var_120_to_fp16_dtype_0 = const()[name = string("op_120_to_fp16_dtype_0"), val = string("fp16")];
-            tensor<fp16, [1, 1, 1, 512]> var_115_to_fp16 = cast(dtype = var_120_to_fp16_dtype_0, x = var_115)[name = string("cast_6")];
-            tensor<fp16, [1, 8, 512, 512]> weights_1_cast_fp16 = mul(x = var_119_cast_fp16, y = var_115_to_fp16)[name = string("weights_1_cast_fp16")];
-            tensor<int32, [1]> var_123_axes_0 = const()[name = string("op_123_axes_0"), val = tensor<int32, [1]>([-1])];
-            bool var_123_keep_dims_0 = const()[name = string("op_123_keep_dims_0"), val = bool(true)];
-            tensor<fp16, [1, 8, 512, 1]> var_123_cast_fp16 = reduce_sum(axes = var_123_axes_0, keep_dims = var_123_keep_dims_0, x = weights_1_cast_fp16)[name = string("op_123_cast_fp16")];
-            fp16 var_21_to_fp16 = const()[name = string("op_21_to_fp16"), val = fp16(0x1.1p-20)];
-            tensor<fp16, [1, 8, 512, 1]> var_124_cast_fp16 = maximum(x = var_123_cast_fp16, y = var_21_to_fp16)[name = string("op_124_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> weights_3_cast_fp16 = real_div(x = weights_1_cast_fp16, y = var_124_cast_fp16)[name = string("weights_3_cast_fp16")];
-            bool var_126_transpose_x_0 = const()[name = string("op_126_transpose_x_0"), val = bool(false)];
-            bool var_126_transpose_y_0 = const()[name = string("op_126_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 512, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = squeeze_2_cast_fp16)[name = string("transpose_17")];
-            tensor<fp16, [1, 8, 512, 64]> var_126_cast_fp16 = matmul(transpose_x = var_126_transpose_x_0, transpose_y = var_126_transpose_y_0, x = weights_3_cast_fp16, y = value_3_cast_fp16)[name = string("op_126_cast_fp16")];
-            tensor<int32, [4]> var_127_perm_0 = const()[name = string("op_127_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
-            tensor<int32, [3]> var_128 = const()[name = string("op_128"), val = tensor<int32, [3]>([1, 512, 512])];
-            tensor<fp16, [1, 512, 8, 64]> var_127_cast_fp16 = transpose(perm = var_127_perm_0, x = var_126_cast_fp16)[name = string("transpose_16")];
-            tensor<fp16, [1, 512, 512]> input_9_cast_fp16 = reshape(shape = var_128, x = var_127_cast_fp16)[name = string("input_9_cast_fp16")];
-            tensor<fp16, [512, 512]> module_flow_span_ordered_encoder_blocks_0_attention_proj_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237888))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1500096))))[name = string("module_flow_span_ordered_encoder_blocks_0_attention_proj_weight_to_fp16_quantized")];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_attention_proj_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_attention_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1501184)))];
-            tensor<fp16, [1, 512, 512]> linear_1_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_0_attention_proj_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_0_attention_proj_weight_to_fp16_quantized, x = input_9_cast_fp16)[name = string("linear_1_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> input_13_cast_fp16 = add(x = hidden_3_cast_fp16, y = linear_1_cast_fp16)[name = string("input_13_cast_fp16")];
-            tensor<int32, [1]> input_15_axes_0 = const()[name = string("input_15_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_norm_feedforward_weight_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_norm_feedforward_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1502272)))];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_norm_feedforward_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_norm_feedforward_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1503360)))];
-            tensor<fp16, [1, 512, 512]> input_15_cast_fp16 = layer_norm(axes = input_15_axes_0, beta = module_flow_span_ordered_encoder_blocks_0_norm_feedforward_bias_to_fp16, epsilon = var_28_to_fp16, gamma = module_flow_span_ordered_encoder_blocks_0_norm_feedforward_weight_to_fp16, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
-            tensor<fp16, [2048, 512]> module_flow_span_ordered_encoder_blocks_0_feedforward_0_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1504448))), scale = tensor<fp16, [2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2553088))))[name = string("module_flow_span_ordered_encoder_blocks_0_feedforward_0_weight_to_fp16_quantized")];
-            tensor<fp16, [2048]> module_flow_span_ordered_encoder_blocks_0_feedforward_0_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_feedforward_0_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2557248)))];
-            tensor<fp16, [1, 512, 2048]> linear_2_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_0_feedforward_0_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_0_feedforward_0_weight_to_fp16_quantized, x = input_15_cast_fp16)[name = string("linear_2_cast_fp16")];
-            string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("TANH_APPROXIMATION")];
-            tensor<fp16, [1, 512, 2048]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_2_cast_fp16)[name = string("input_19_cast_fp16")];
-            tensor<fp16, [512, 2048]> module_flow_span_ordered_encoder_blocks_0_feedforward_2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2561408))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3610048))))[name = string("module_flow_span_ordered_encoder_blocks_0_feedforward_2_weight_to_fp16_quantized")];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_0_feedforward_2_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_0_feedforward_2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3611136)))];
-            tensor<fp16, [1, 512, 512]> linear_3_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_0_feedforward_2_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_0_feedforward_2_weight_to_fp16_quantized, x = input_19_cast_fp16)[name = string("linear_3_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> hidden_5_cast_fp16 = add(x = input_13_cast_fp16, y = linear_3_cast_fp16)[name = string("hidden_5_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> hidden_7_cast_fp16 = select(a = var_18_to_fp16, b = hidden_5_cast_fp16, cond = var_83)[name = string("hidden_7_cast_fp16")];
-            tensor<int32, [1]> input_23_axes_0 = const()[name = string("input_23_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_norm_attention_weight_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_norm_attention_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3612224)))];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_norm_attention_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_norm_attention_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3613312)))];
-            tensor<fp16, [1, 512, 512]> input_23_cast_fp16 = layer_norm(axes = input_23_axes_0, beta = module_flow_span_ordered_encoder_blocks_1_norm_attention_bias_to_fp16, epsilon = var_28_to_fp16, gamma = module_flow_span_ordered_encoder_blocks_1_norm_attention_weight_to_fp16, x = hidden_7_cast_fp16)[name = string("input_23_cast_fp16")];
-            tensor<fp16, [1536, 512]> module_flow_span_ordered_encoder_blocks_1_qkv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3614400))), scale = tensor<fp16, [1536, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4400896))))[name = string("module_flow_span_ordered_encoder_blocks_1_qkv_weight_to_fp16_quantized")];
-            tensor<fp16, [1536]> module_flow_span_ordered_encoder_blocks_1_qkv_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_qkv_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4404032)))];
-            tensor<fp16, [1, 512, 1536]> linear_4_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_1_qkv_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_1_qkv_weight_to_fp16_quantized, x = input_23_cast_fp16)[name = string("linear_4_cast_fp16")];
-            tensor<int32, [5]> var_168 = const()[name = string("op_168"), val = tensor<int32, [5]>([1, 512, 3, 8, 64])];
-            tensor<fp16, [1, 512, 3, 8, 64]> qkv_3_cast_fp16 = reshape(shape = var_168, x = linear_4_cast_fp16)[name = string("qkv_3_cast_fp16")];
-            tensor<int32, [3]> var_170_split_sizes_0 = const()[name = string("op_170_split_sizes_0"), val = tensor<int32, [3]>([1, 1, 1])];
-            int32 var_170_axis_0 = const()[name = string("op_170_axis_0"), val = int32(2)];
-            tensor<fp16, [1, 512, 1, 8, 64]> var_170_cast_fp16_0, tensor<fp16, [1, 512, 1, 8, 64]> var_170_cast_fp16_1, tensor<fp16, [1, 512, 1, 8, 64]> var_170_cast_fp16_2 = split(axis = var_170_axis_0, split_sizes = var_170_split_sizes_0, x = qkv_3_cast_fp16)[name = string("op_170_cast_fp16")];
-            tensor<int32, [1]> squeeze_3_axes_0 = const()[name = string("squeeze_3_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_3_cast_fp16 = squeeze(axes = squeeze_3_axes_0, x = var_170_cast_fp16_0)[name = string("squeeze_3_cast_fp16")];
-            tensor<int32, [1]> squeeze_4_axes_0 = const()[name = string("squeeze_4_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_4_cast_fp16 = squeeze(axes = squeeze_4_axes_0, x = var_170_cast_fp16_1)[name = string("squeeze_4_cast_fp16")];
-            tensor<int32, [1]> squeeze_5_axes_0 = const()[name = string("squeeze_5_axes_0"), val = tensor<int32, [1]>([2])];
-            tensor<fp16, [1, 512, 8, 64]> squeeze_5_cast_fp16 = squeeze(axes = squeeze_5_axes_0, x = var_170_cast_fp16_2)[name = string("squeeze_5_cast_fp16")];
-            tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            bool var_178_transpose_x_0 = const()[name = string("op_178_transpose_x_0"), val = bool(false)];
-            bool var_178_transpose_y_0 = const()[name = string("op_178_transpose_y_0"), val = bool(false)];
-            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
-            tensor<int32, [4]> transpose_11_perm_0 = const()[name = string("transpose_11_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
-            tensor<fp16, [1, 8, 64, 512]> transpose_11 = transpose(perm = transpose_11_perm_0, x = squeeze_4_cast_fp16)[name = string("transpose_14")];
-            tensor<fp16, [1, 8, 512, 64]> transpose_10 = transpose(perm = transpose_10_perm_0, x = squeeze_3_cast_fp16)[name = string("transpose_15")];
-            tensor<fp16, [1, 8, 512, 512]> var_178_cast_fp16 = matmul(transpose_x = var_178_transpose_x_0, transpose_y = var_178_transpose_y_0, x = transpose_10, y = transpose_11)[name = string("op_178_cast_fp16")];
-            fp16 var_179_to_fp16 = const()[name = string("op_179_to_fp16"), val = fp16(0x1p-3)];
-            tensor<fp16, [1, 8, 512, 512]> scores_5_cast_fp16 = mul(x = var_178_cast_fp16, y = var_179_to_fp16)[name = string("scores_5_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> scores_cast_fp16 = select(a = var_23_to_fp16, b = scores_5_cast_fp16, cond = var_117)[name = string("scores_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> var_187_cast_fp16 = softmax(axis = var_30, x = scores_cast_fp16)[name = string("op_187_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> weights_5_cast_fp16 = mul(x = var_187_cast_fp16, y = var_115_to_fp16)[name = string("weights_5_cast_fp16")];
-            tensor<int32, [1]> var_191_axes_0 = const()[name = string("op_191_axes_0"), val = tensor<int32, [1]>([-1])];
-            bool var_191_keep_dims_0 = const()[name = string("op_191_keep_dims_0"), val = bool(true)];
-            tensor<fp16, [1, 8, 512, 1]> var_191_cast_fp16 = reduce_sum(axes = var_191_axes_0, keep_dims = var_191_keep_dims_0, x = weights_5_cast_fp16)[name = string("op_191_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 1]> var_192_cast_fp16 = maximum(x = var_191_cast_fp16, y = var_21_to_fp16)[name = string("op_192_cast_fp16")];
-            tensor<fp16, [1, 8, 512, 512]> weights_cast_fp16 = real_div(x = weights_5_cast_fp16, y = var_192_cast_fp16)[name = string("weights_cast_fp16")];
-            bool var_194_transpose_x_0 = const()[name = string("op_194_transpose_x_0"), val = bool(false)];
-            bool var_194_transpose_y_0 = const()[name = string("op_194_transpose_y_0"), val = bool(false)];
-            tensor<fp16, [1, 8, 512, 64]> value_cast_fp16 = transpose(perm = value_perm_0, x = squeeze_5_cast_fp16)[name = string("transpose_13")];
-            tensor<fp16, [1, 8, 512, 64]> var_194_cast_fp16 = matmul(transpose_x = var_194_transpose_x_0, transpose_y = var_194_transpose_y_0, x = weights_cast_fp16, y = value_cast_fp16)[name = string("op_194_cast_fp16")];
-            tensor<int32, [4]> var_195_perm_0 = const()[name = string("op_195_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
-            tensor<int32, [3]> var_196 = const()[name = string("op_196"), val = tensor<int32, [3]>([1, 512, 512])];
-            tensor<fp16, [1, 512, 8, 64]> var_195_cast_fp16 = transpose(perm = var_195_perm_0, x = var_194_cast_fp16)[name = string("transpose_12")];
-            tensor<fp16, [1, 512, 512]> input_25_cast_fp16 = reshape(shape = var_196, x = var_195_cast_fp16)[name = string("input_25_cast_fp16")];
-            tensor<fp16, [512, 512]> module_flow_span_ordered_encoder_blocks_1_attention_proj_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4407168))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4669376))))[name = string("module_flow_span_ordered_encoder_blocks_1_attention_proj_weight_to_fp16_quantized")];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_attention_proj_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_attention_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4670464)))];
-            tensor<fp16, [1, 512, 512]> linear_5_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_1_attention_proj_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_1_attention_proj_weight_to_fp16_quantized, x = input_25_cast_fp16)[name = string("linear_5_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> input_29_cast_fp16 = add(x = hidden_7_cast_fp16, y = linear_5_cast_fp16)[name = string("input_29_cast_fp16")];
-            tensor<int32, [1]> input_31_axes_0 = const()[name = string("input_31_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_norm_feedforward_weight_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_norm_feedforward_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4671552)))];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_norm_feedforward_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_norm_feedforward_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4672640)))];
-            tensor<fp16, [1, 512, 512]> input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = module_flow_span_ordered_encoder_blocks_1_norm_feedforward_bias_to_fp16, epsilon = var_28_to_fp16, gamma = module_flow_span_ordered_encoder_blocks_1_norm_feedforward_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")];
-            tensor<fp16, [2048, 512]> module_flow_span_ordered_encoder_blocks_1_feedforward_0_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4673728))), scale = tensor<fp16, [2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5722368))))[name = string("module_flow_span_ordered_encoder_blocks_1_feedforward_0_weight_to_fp16_quantized")];
-            tensor<fp16, [2048]> module_flow_span_ordered_encoder_blocks_1_feedforward_0_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_feedforward_0_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5726528)))];
-            tensor<fp16, [1, 512, 2048]> linear_6_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_1_feedforward_0_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_1_feedforward_0_weight_to_fp16_quantized, x = input_31_cast_fp16)[name = string("linear_6_cast_fp16")];
-            string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("TANH_APPROXIMATION")];
-            tensor<fp16, [1, 512, 2048]> input_35_cast_fp16 = gelu(mode = input_35_mode_0, x = linear_6_cast_fp16)[name = string("input_35_cast_fp16")];
-            tensor<fp16, [512, 2048]> module_flow_span_ordered_encoder_blocks_1_feedforward_2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5730688))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6779328))))[name = string("module_flow_span_ordered_encoder_blocks_1_feedforward_2_weight_to_fp16_quantized")];
-            tensor<fp16, [512]> module_flow_span_ordered_encoder_blocks_1_feedforward_2_bias_to_fp16 = const()[name = string("module_flow_span_ordered_encoder_blocks_1_feedforward_2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6780416)))];
-            tensor<fp16, [1, 512, 512]> linear_7_cast_fp16 = linear(bias = module_flow_span_ordered_encoder_blocks_1_feedforward_2_bias_to_fp16, weight = module_flow_span_ordered_encoder_blocks_1_feedforward_2_weight_to_fp16_quantized, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> hidden_9_cast_fp16 = add(x = input_29_cast_fp16, y = linear_7_cast_fp16)[name = string("hidden_9_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> hidden_cast_fp16 = select(a = var_18_to_fp16, b = hidden_9_cast_fp16, cond = var_83)[name = string("hidden_cast_fp16")];
-            tensor<bool, [1, 512]> var_224 = equal(x = input_3, y = var_40)[name = string("op_224")];
-            tensor<bool, [1, 512]> var_225 = logical_and(x = span_context_mask_1, y = var_224)[name = string("op_225")];
-            tensor<int32, [1]> var_228_axes_0 = const()[name = string("op_228_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_228_keep_dims_0 = const()[name = string("op_228_keep_dims_0"), val = bool(true)];
-            string weight_15_to_fp16_dtype_0 = const()[name = string("weight_15_to_fp16_dtype_0"), val = string("fp16")];
-            tensor<fp16, [1, 512]> var_225_to_fp16 = cast(dtype = weight_15_to_fp16_dtype_0, x = var_225)[name = string("cast_5")];
-            tensor<fp16, [1, 1]> var_228_cast_fp16 = reduce_sum(axes = var_228_axes_0, keep_dims = var_228_keep_dims_0, x = var_225_to_fp16)[name = string("op_228_cast_fp16")];
-            fp16 var_17_to_fp16 = const()[name = string("op_17_to_fp16"), val = fp16(0x1p+0)];
-            tensor<fp16, [1, 1]> var_229_cast_fp16 = maximum(x = var_228_cast_fp16, y = var_17_to_fp16)[name = string("op_229_cast_fp16")];
-            tensor<fp16, [1, 512]> weight_17_cast_fp16 = real_div(x = var_225_to_fp16, y = var_229_cast_fp16)[name = string("weight_17_cast_fp16")];
-            tensor<int32, [1]> var_231_axes_0 = const()[name = string("op_231_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1, 512, 1]> var_231_cast_fp16 = expand_dims(axes = var_231_axes_0, x = weight_17_cast_fp16)[name = string("op_231_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> var_232_cast_fp16 = mul(x = hidden_cast_fp16, y = var_231_cast_fp16)[name = string("op_232_cast_fp16")];
-            tensor<int32, [1]> var_234_axes_0 = const()[name = string("op_234_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_234_keep_dims_0 = const()[name = string("op_234_keep_dims_0"), val = bool(false)];
-            tensor<fp16, [1, 512]> var_234_cast_fp16 = reduce_sum(axes = var_234_axes_0, keep_dims = var_234_keep_dims_0, x = var_232_cast_fp16)[name = string("op_234_cast_fp16")];
-            tensor<bool, [1, 512]> var_235 = equal(x = input_3, y = var_38)[name = string("op_235")];
-            tensor<bool, [1, 512]> var_236 = logical_and(x = span_context_mask_1, y = var_235)[name = string("op_236")];
-            tensor<int32, [1]> var_239_axes_0 = const()[name = string("op_239_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_239_keep_dims_0 = const()[name = string("op_239_keep_dims_0"), val = bool(true)];
-            string weight_19_to_fp16_dtype_0 = const()[name = string("weight_19_to_fp16_dtype_0"), val = string("fp16")];
-            tensor<fp16, [1, 512]> var_236_to_fp16 = cast(dtype = weight_19_to_fp16_dtype_0, x = var_236)[name = string("cast_4")];
-            tensor<fp16, [1, 1]> var_239_cast_fp16 = reduce_sum(axes = var_239_axes_0, keep_dims = var_239_keep_dims_0, x = var_236_to_fp16)[name = string("op_239_cast_fp16")];
-            tensor<fp16, [1, 1]> var_240_cast_fp16 = maximum(x = var_239_cast_fp16, y = var_17_to_fp16)[name = string("op_240_cast_fp16")];
-            tensor<fp16, [1, 512]> weight_21_cast_fp16 = real_div(x = var_236_to_fp16, y = var_240_cast_fp16)[name = string("weight_21_cast_fp16")];
-            tensor<int32, [1]> var_242_axes_0 = const()[name = string("op_242_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1, 512, 1]> var_242_cast_fp16 = expand_dims(axes = var_242_axes_0, x = weight_21_cast_fp16)[name = string("op_242_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> var_243_cast_fp16 = mul(x = hidden_cast_fp16, y = var_242_cast_fp16)[name = string("op_243_cast_fp16")];
-            tensor<int32, [1]> var_245_axes_0 = const()[name = string("op_245_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_245_keep_dims_0 = const()[name = string("op_245_keep_dims_0"), val = bool(false)];
-            tensor<fp16, [1, 512]> var_245_cast_fp16 = reduce_sum(axes = var_245_axes_0, keep_dims = var_245_keep_dims_0, x = var_243_cast_fp16)[name = string("op_245_cast_fp16")];
-            tensor<bool, [1, 512]> var_246 = equal(x = input_3, y = var_35)[name = string("op_246")];
-            tensor<bool, [1, 512]> var_247 = logical_and(x = span_context_mask_1, y = var_246)[name = string("op_247")];
-            tensor<int32, [1]> var_250_axes_0 = const()[name = string("op_250_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_250_keep_dims_0 = const()[name = string("op_250_keep_dims_0"), val = bool(true)];
-            string weight_23_to_fp16_dtype_0 = const()[name = string("weight_23_to_fp16_dtype_0"), val = string("fp16")];
-            tensor<fp16, [1, 512]> var_247_to_fp16 = cast(dtype = weight_23_to_fp16_dtype_0, x = var_247)[name = string("cast_3")];
-            tensor<fp16, [1, 1]> var_250_cast_fp16 = reduce_sum(axes = var_250_axes_0, keep_dims = var_250_keep_dims_0, x = var_247_to_fp16)[name = string("op_250_cast_fp16")];
-            tensor<fp16, [1, 1]> var_251_cast_fp16 = maximum(x = var_250_cast_fp16, y = var_17_to_fp16)[name = string("op_251_cast_fp16")];
-            tensor<fp16, [1, 512]> weight_25_cast_fp16 = real_div(x = var_247_to_fp16, y = var_251_cast_fp16)[name = string("weight_25_cast_fp16")];
-            tensor<int32, [1]> var_253_axes_0 = const()[name = string("op_253_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [1, 512, 1]> var_253_cast_fp16 = expand_dims(axes = var_253_axes_0, x = weight_25_cast_fp16)[name = string("op_253_cast_fp16")];
-            tensor<fp16, [1, 512, 512]> var_254_cast_fp16 = mul(x = hidden_cast_fp16, y = var_253_cast_fp16)[name = string("op_254_cast_fp16")];
-            tensor<int32, [1]> var_256_axes_0 = const()[name = string("op_256_axes_0"), val = tensor<int32, [1]>([1])];
-            bool var_256_keep_dims_0 = const()[name = string("op_256_keep_dims_0"), val = bool(false)];
-            tensor<fp16, [1, 512]> var_256_cast_fp16 = reduce_sum(axes = var_256_axes_0, keep_dims = var_256_keep_dims_0, x = var_254_cast_fp16)[name = string("op_256_cast_fp16")];
-            bool input_39_interleave_0 = const()[name = string("input_39_interleave_0"), val = bool(false)];
-            tensor<fp16, [1, 1536]> input_39_cast_fp16 = concat(axis = var_30, interleave = input_39_interleave_0, values = (var_234_cast_fp16, var_245_cast_fp16, var_256_cast_fp16))[name = string("input_39_cast_fp16")];
-            tensor<fp16, [512, 1536]> module_flow_span_segment_pool_proj_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6781504))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7568000))))[name = string("module_flow_span_segment_pool_proj_weight_to_fp16_quantized")];
-            tensor<fp16, [512]> module_flow_span_segment_pool_proj_bias_to_fp16 = const()[name = string("module_flow_span_segment_pool_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7569088)))];
-            tensor<fp16, [1, 512]> linear_8_cast_fp16 = linear(bias = module_flow_span_segment_pool_proj_bias_to_fp16, weight = module_flow_span_segment_pool_proj_weight_to_fp16_quantized, x = input_39_cast_fp16)[name = string("linear_8_cast_fp16")];
-            tensor<int32, [1]> pooled_axes_0 = const()[name = string("pooled_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp16, [512]> module_flow_span_context_norm_weight_to_fp16 = const()[name = string("module_flow_span_context_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7570176)))];
-            tensor<fp16, [512]> module_flow_span_context_norm_bias_to_fp16 = const()[name = string("module_flow_span_context_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7571264)))];
-            tensor<fp16, [1, 512]> pooled_cast_fp16 = layer_norm(axes = pooled_axes_0, beta = module_flow_span_context_norm_bias_to_fp16, epsilon = var_28_to_fp16, gamma = module_flow_span_context_norm_weight_to_fp16, x = linear_8_cast_fp16)[name = string("pooled_cast_fp16")];
-            tensor<int32, [1]> var_38_list = const()[name = string("op_38_list"), val = tensor<int32, [1]>([1])];
-            bool reduce_max_0_keep_dims_0 = const()[name = string("reduce_max_0_keep_dims_0"), val = bool(true)];
-            tensor<int32, [1, 1]> reduce_max_0 = reduce_max(axes = var_38_list, keep_dims = reduce_max_0_keep_dims_0, x = var_64)[name = string("reduce_max_0")];
-            string var_266_dtype_0 = const()[name = string("op_266_dtype_0"), val = string("bool")];
-            tensor<bool, [1, 1]> var_266 = cast(dtype = var_266_dtype_0, x = reduce_max_0)[name = string("cast_2")];
-            tensor<bool, [1, 1]> var_267 = logical_not(x = var_266)[name = string("op_267")];
-            tensor<fp16, [1, 512]> var_268_cast_fp16 = select(a = var_18_to_fp16, b = pooled_cast_fp16, cond = var_267)[name = string("op_268_cast_fp16")];
-            string var_268_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_268_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
-            tensor<fp32, [1, 512]> span_context_hidden = cast(dtype = var_268_cast_fp16_to_fp32_dtype_0, x = var_268_cast_fp16)[name = string("cast_1")];
-        } -> (span_context_hidden);
-}

coreml/coreml/vector_context_encoder.mlmodelc/weights/weight.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:70da5deb1236e2f5d3250cef65ee0619f1089ebb7c5fc39744d205d45bf24c66
-size 7572352

coreml/coreml/vector_estimator.mlmodelc/analytics/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:13eab20d76b95b33201ab96dd522599eabd282b36db8c231832834862d4385b7
-size 243

coreml/coreml/vector_estimator.mlmodelc/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d22410a5a41fc0aae79224588c2da505cd4abae60526d06fd989749001741762
-size 3647

coreml/coreml/vector_estimator.mlmodelc/model.mil DELETED Viewed

The diff for this file is too large to render. See raw diff

coreml/coreml/vector_estimator.mlmodelc/weights/weight.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:dfcd7719b9406f6a181f17f48125c33213196f9e722844f88ecbae5a580a8015
-size 60754048

coreml/coreml/vocoder.mlmodelc/analytics/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1265ad8d6e6ff623afa255fc3eab94c840e6b372ec6d2ebcacfe9ee053c9aade
-size 243

coreml/coreml/vocoder.mlmodelc/coremldata.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:04bf9f9de7edfc6ce58e5928a0a1d6669d09095ebc03cb5ca24dd85a416653df
-size 1190

coreml/coreml/vocoder.mlmodelc/model.mil DELETED Viewed

The diff for this file is too large to render. See raw diff

coreml/coreml/vocoder.mlmodelc/weights/weight.bin DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8d1aa1e85be960d5ad32f751b602c38601bc9517a951be5bd42c991449d3bccc
-size 31043584

coreml/manifest.json DELETED Viewed

@@ -1,998 +0,0 @@
-{
-  "architecture": "scyllasband-duration-flow",
-  "assets": {
-    "fixed_delivery_references": "assets/fixed_references.json",
-    "g2p_config": "assets/g2p/export_config.json",
-    "g2p_language_map": "assets/g2p/language_map.json",
-    "g2p_normalization": "assets/g2p/normalization.json",
-    "g2p_phoneme_dict": "assets/g2p/phoneme_dict.json",
-    "g2p_pronunciation_overrides": "assets/g2p/pronunciation_overrides.json",
-    "g2p_tokenizer": "assets/g2p/tokenizer.json",
-    "language_index": "assets/languages.json",
-    "phone_vocab": "assets/phone_vocab.json",
-    "voice_index": "assets/voices.json"
-  },
-  "audio": {
-    "hop_length": 256,
-    "latent_dim": 24,
-    "latent_hop_length": 512,
-    "n_mels": 100,
-    "sample_rate": 24000
-  },
-  "components": {
-    "g2p_64": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/g2p.mlmodelc",
-          "function": "g2p_64"
-        }
-      },
-      "inputs": [
-        "text"
-      ],
-      "outputs": [
-        "logits"
-      ],
-      "required": false
-    },
-    "g2p_128": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/g2p.mlmodelc",
-          "function": "g2p_128"
-        }
-      },
-      "inputs": [
-        "text"
-      ],
-      "outputs": [
-        "logits"
-      ],
-      "required": false
-    },
-    "g2p_256": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/g2p.mlmodelc",
-          "function": "g2p_256"
-        }
-      },
-      "inputs": [
-        "text"
-      ],
-      "outputs": [
-        "logits"
-      ],
-      "required": false
-    },
-    "g2p": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/g2p.mlmodelc",
-          "function": "g2p"
-        }
-      },
-      "inputs": [
-        "text"
-      ],
-      "outputs": [
-        "logits"
-      ],
-      "required": true
-    },
-    "duration_predictor": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/duration_predictor.mlmodelc"
-        }
-      },
-      "inputs": [
-        "span_phone_ids",
-        "span_segment_ids",
-        "span_mask",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present"
-      ],
-      "outputs": [
-        "durations"
-      ],
-      "required": true
-    },
-    "vector_context_encoder": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_context_encoder.mlmodelc"
-        }
-      },
-      "inputs": [
-        "span_context_phone_ids",
-        "span_context_segment_ids",
-        "span_context_mask"
-      ],
-      "outputs": [
-        "span_context_hidden"
-      ],
-      "required": true
-    },
-    "vector_estimator_128": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator_128"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": false
-    },
-    "vocoder_128": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder_128"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": false
-    },
-    "vector_estimator_256": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator_256"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": false
-    },
-    "vocoder_256": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder_256"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": false
-    },
-    "vector_estimator_384": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator_384"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": false
-    },
-    "vocoder_384": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder_384"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": false
-    },
-    "vector_estimator_512": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator_512"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": false
-    },
-    "vocoder_512": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder_512"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": false
-    },
-    "vector_estimator_768": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator_768"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": false
-    },
-    "vocoder_768": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder_768"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": false
-    },
-    "vector_estimator": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vector_estimator.mlmodelc",
-          "function": "vector_estimator"
-        }
-      },
-      "inputs": [
-        "noise",
-        "time",
-        "expanded_phone_ids",
-        "voice_id",
-        "language_id",
-        "delivery_values",
-        "delivery_present",
-        "latent_mask",
-        "span_context_hidden",
-        "prefix_latents",
-        "prefix_mask",
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "outputs": [
-        "velocity"
-      ],
-      "required": true
-    },
-    "vocoder": {
-      "artifacts": {
-        "coreml": {
-          "format": "coreml",
-          "path": "coreml/vocoder.mlmodelc",
-          "function": "vocoder"
-        }
-      },
-      "inputs": [
-        "latents",
-        "latent_mask",
-        "voice_id",
-        "language_id",
-        "emotion_id"
-      ],
-      "outputs": [
-        "audio"
-      ],
-      "required": true
-    }
-  },
-  "contract_version": "1.0.0",
-  "controls": {
-    "chunk_context": {
-      "artificial_continuation_boundaries": [
-        "chunk_continue"
-      ],
-      "artificial_continuation_phone": null,
-      "boundary_after_values": [
-        "sentence_end",
-        "paragraph_end",
-        "clause_continue",
-        "chunk_continue"
-      ],
-      "boundary_before_values": [
-        "sentence_start",
-        "paragraph_start",
-        "clause_continue",
-        "chunk_continue"
-      ],
-      "clause_continuation_fallback_phone": "<pause_comma>",
-      "conditioning": "boundary_embeddings",
-      "context_phone_tokens": {
-        "continuation": "<ctx_continuation>",
-        "sentence_end": "<ctx_sentence_end>",
-        "sentence_start": "<ctx_sentence_start>"
-      },
-      "default_boundary_after": "sentence_end",
-      "default_boundary_before": "sentence_start",
-      "emit_context_phone_tokens": false,
-      "linguistic_continuation_boundaries": [
-        "clause_continue"
-      ],
-      "schema_version": 2
-    },
-    "chunking": {
-      "shared_split_phone": "<sil>",
-      "split_after": [
-        "<end_stmt>",
-        "<end_question>",
-        "<end_exclaim>",
-        "<ellipsis>"
-      ],
-      "unit": "sentence"
-    },
-    "decoding": {
-      "context_frames": 48,
-      "mode": "overlap"
-    },
-    "delivery": {
-      "axes": [
-        "energy",
-        "tension",
-        "valence",
-        "assertiveness"
-      ],
-      "default": "neutral",
-      "enabled": true,
-      "neutral": 2.0,
-      "normalization": "(rating-2)/4",
-      "presence_mask": true,
-      "range": [
-        0.0,
-        4.0
-      ],
-      "release_id": "v2-20261006",
-      "scope": "utterance",
-      "timeline_conditioning": false,
-      "voice_locales": {
-        "ariadne": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "felix": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "gwen": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "ink": [
-          "de",
-          "en_gb",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "max": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "orpheus": [
-          "de",
-          "en_gb",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "rex": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "scylla": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "stone": [
-          "de",
-          "en_us",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ],
-        "tuesday": [
-          "de",
-          "en_gb",
-          "es",
-          "fr",
-          "it",
-          "vi"
-        ]
-      }
-    },
-    "durations": {
-      "rounding": "half_even",
-      "silence_min_frames": 0,
-      "speech_min_frames": 1,
-      "zero_frame_phones": "punctuation_and_modifiers"
-    },
-    "emotions": {
-      "enabled": false
-    },
-    "energy": {
-      "default": 1.0,
-      "max": 1.5,
-      "min": 0.5
-    },
-    "fixed_shapes": {
-      "g2p_text_tokens": 512,
-      "latent_frames": 1024,
-      "phone_frames": 512
-    },
-    "graph_input_contract": "scyllasband_measured_delivery_v2",
-    "pitch": {
-      "default": 0.0,
-      "max": 4.0,
-      "min": -4.0
-    },
-    "prefix_conditioning": {
-      "alignment": "right",
-      "enabled": true,
-      "max_frames": 96,
-      "source": "previous_latents_tail"
-    },
-    "punctuation_silence": {
-      "calibrated_floors": false,
-      "floor_table_ms": {},
-      "silence_phone": "<sil>",
-      "target": "explicit_silence"
-    },
-    "reference_packs": {
-      "enabled": false,
-      "fixed_in_graph": true
-    },
-    "sampler": {
-      "choices": [
-        "euler",
-        "heun"
-      ],
-      "default": "euler"
-    },
-    "span_conditioning": {
-      "context_max_phones": 512,
-      "context_phones_each_side": 180,
-      "duration_inputs": [
-        "span_phone_ids",
-        "span_segment_ids",
-        "span_mask"
-      ],
-      "enabled": true,
-      "encoder_component": "vector_context_encoder",
-      "encoder_inputs": [
-        "span_context_phone_ids",
-        "span_context_segment_ids",
-        "span_context_mask"
-      ],
-      "exclude_context_phones": [
-        "<sil>"
-      ],
-      "hidden_size": 512,
-      "layout": "balanced_before_target_after",
-      "scope": "passage",
-      "segments": 3,
-      "solver_input": "span_context_hidden"
-    },
-    "speed": {
-      "default": 1.0,
-      "max": 1.25,
-      "min": 0.75
-    },
-    "steps": {
-      "default": 4,
-      "max": 64,
-      "min": 1
-    },
-    "target_buckets": {
-      "bucket_margin_frames": 0,
-      "buckets": [
-        {
-          "default": false,
-          "latent_frames": 128,
-          "vector_estimator": "vector_estimator_128",
-          "vocoder": "vocoder_128"
-        },
-        {
-          "default": false,
-          "latent_frames": 256,
-          "vector_estimator": "vector_estimator_256",
-          "vocoder": "vocoder_256"
-        },
-        {
-          "default": false,
-          "latent_frames": 384,
-          "vector_estimator": "vector_estimator_384",
-          "vocoder": "vocoder_384"
-        },
-        {
-          "default": false,
-          "latent_frames": 512,
-          "vector_estimator": "vector_estimator_512",
-          "vocoder": "vocoder_512"
-        },
-        {
-          "default": false,
-          "latent_frames": 768,
-          "vector_estimator": "vector_estimator_768",
-          "vocoder": "vocoder_768"
-        },
-        {
-          "default": true,
-          "latent_frames": 1024,
-          "vector_estimator": "vector_estimator",
-          "vocoder": "vocoder"
-        }
-      ],
-      "default_latent_frames": 1024,
-      "enabled": true,
-      "policy": "smallest_fit"
-    },
-    "text_normalizer": {
-      "prepared_contract_recorded": true,
-      "sha256": "d8fd1b2415d0d100d1c70237a4469117ee6b73d0e4da703e4654c83192779cc4",
-      "version": "scyllasband_spoken_text_v2_2026_08_05"
-    },
-    "timing_events": {
-      "enabled": true,
-      "inputs": [
-        "expanded_boundary_event_ids",
-        "expanded_modifier_event_ids",
-        "expanded_phone_phase",
-        "expanded_phone_log_duration",
-        "expanded_sentence_type_ids"
-      ],
-      "modifier_bits": {
-        "ˈ": 1,
-        "ˌ": 2,
-        "ː": 4,
-        "̃": 8,
-        "̩": 16,
-        "̪": 32
-      },
-      "punctuation_owner": "following_silence_else_first_frame_of_next_phone",
-      "sentence_type_count": 5,
-      "sentence_type_lookahead": "following_context",
-      "sentence_types": {
-        "<ellipsis>": 4,
-        "<end_exclaim>": 3,
-        "<end_question>": 2,
-        "<end_stmt>": 1
-      },
-      "word_start_event_id": 87,
-      "word_start_owner": "first_phone_of_word_after_chunk_first_word"
-    },
-    "word_boundaries": {
-      "duration_phone": "<sil>",
-      "enabled": true,
-      "g2p_output_symbol": " ",
-      "zero_frames_allowed": true
-    },
-    "g2p_buckets": [
-      {
-        "text_tokens": 64,
-        "component": "g2p_64"
-      },
-      {
-        "text_tokens": 128,
-        "component": "g2p_128"
-      },
-      {
-        "text_tokens": 256,
-        "component": "g2p_256"
-      },
-      {
-        "text_tokens": 512,
-        "component": "g2p"
-      }
-    ],
-    "coreml": {
-      "minimum_os": {
-        "ios": "18.0",
-        "macos": "15.0"
-      },
-      "io": {
-        "ids": "int32",
-        "masks": "int32",
-        "values": "float32"
-      },
-      "assets": {
-        "g2p": {
-          "precision": "fp16",
-          "weights": "w8",
-          "bytes": 27735781
-        },
-        "duration_predictor": {
-          "precision": "fp16",
-          "weights": "w8",
-          "bytes": 10702662
-        },
-        "vector_context_encoder": {
-          "precision": "fp16",
-          "weights": "w8",
-          "bytes": 7617880
-        },
-        "vector_estimator": {
-          "precision": "fp16",
-          "weights": "w8",
-          "bytes": 62226942
-        },
-        "vocoder": {
-          "precision": "fp16",
-          "weights": "w8",
-          "bytes": 32014276
-        }
-      },
-      "compute_units": "gpu",
-      "neural_engine_assets": []
-    }
-  },
-  "dataset_summary": {
-    "controls": "measured_four_axes",
-    "voice_locale_pairs": 60
-  },
-  "default_language": "en_us",
-  "languages": [
-    "en_us",
-    "en_gb",
-    "es",
-    "it",
-    "fr",
-    "de",
-    "vi"
-  ],
-  "model_name": "Scylla's Band",
-  "model_version": "2",
-  "preferred_backends": [
-    "coreml"
-  ],
-  "voices": [
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "ariadne",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "ariadne",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "felix",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "felix",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "gwen",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "gwen",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_gb",
-      "embedding_path": null,
-      "id": "ink",
-      "languages": [
-        "de",
-        "en_gb",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "ink",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "max",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "max",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_gb",
-      "embedding_path": null,
-      "id": "orpheus",
-      "languages": [
-        "de",
-        "en_gb",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "orpheus",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "rex",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "rex",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "scylla",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "scylla",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_us",
-      "embedding_path": null,
-      "id": "stone",
-      "languages": [
-        "de",
-        "en_us",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "stone",
-      "styles": [
-        "default"
-      ]
-    },
-    {
-      "default_language": "en_gb",
-      "embedding_path": null,
-      "id": "tuesday",
-      "languages": [
-        "de",
-        "en_gb",
-        "es",
-        "fr",
-        "it",
-        "vi"
-      ],
-      "name": "tuesday",
-      "styles": [
-        "default"
-      ]
-    }
-  ]
-}

litert/assets/fixed_references.json CHANGED Viewed

The diff for this file is too large to render. See raw diff

litert/assets/voices.json CHANGED Viewed

@@ -38,5 +38,21 @@
   {
     "id": "tuesday",
     "index": 9
   }
 ]

   {
     "id": "tuesday",
     "index": 9
+  },
+  {
+    "id": "thistle",
+    "index": 10
+  },
+  {
+    "id": "kincade",
+    "index": 11
+  },
+  {
+    "id": "poppy",
+    "index": 12
+  },
+  {
+    "id": "priya",
+    "index": 13
   }
 ]

litert/litert/duration_predictor.tflite CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7b063e6121ca62bc7e4940822d1016c47999c1688bd7482b27cd7396d5b5bab
-size 11599200

 version https://git-lfs.github.com/spec/v1
+oid sha256:02d815abcae4b3915f7569a1796364a5b7d70adc8e928fc2b7beb33566f6b444
+size 11668736

litert/litert/vector_context_encoder.tflite CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7d653507b92d8a5102f4281361e8ee8f03cf001b6889e1434ad1e760fbcf057
-size 7763584

 version https://git-lfs.github.com/spec/v1
+oid sha256:de37ff6548202785b9a7c3edac609b9d94d11f437cc959bcfff61fc909272e5e
+size 11894656

litert/litert/vector_estimator.tflite CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51f6ce100d8565fa7979718ca34ad0b19d1e14f6525c41c46112dfbd1d5b30cd
-size 76550528

 version https://git-lfs.github.com/spec/v1
+oid sha256:cebf54a5b5009c4bf47982cf370a61b36bf3279ea8c16f7c0d974a4e5534ed56
+size 146540928

litert/litert/vocoder.tflite CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0410221c7742098fa5efea7f6f7626e522d3a761d84d8bf2e811c9d0de534f3
-size 38613392

 version https://git-lfs.github.com/spec/v1
+oid sha256:bc04828ad6575713d597585b117081c2f0af6185b6ea9eee6c979dd6b94cbf74
+size 38615216

litert/manifest.json CHANGED Viewed

@@ -447,7 +447,7 @@
         0.0,
         4.0
       ],
-      "release_id": "v2-20261006",
       "scope": "utterance",
       "timeline_conditioning": false,
       "voice_locales": {
@@ -483,6 +483,14 @@
           "it",
           "vi"
         ],
         "max": [
           "de",
           "en_us",
@@ -499,6 +507,22 @@
           "it",
           "vi"
         ],
         "rex": [
           "de",
           "en_us",
@@ -523,6 +547,14 @@
           "it",
           "vi"
         ],
         "tuesday": [
           "de",
           "en_gb",
@@ -557,19 +589,19 @@
       "quantization": "dynamic_wi8_afp32",
       "report": {
         "duration_predictor": {
-          "bytes": 11599200,
           "quantized": "dynamic_wi8_afp32"
         },
         "vector_context_encoder": {
-          "bytes": 7763584,
           "quantized": "dynamic_wi8_afp32"
         },
         "vector_estimator": {
-          "bytes": 76550528,
           "quantized": "dynamic_wi8_afp32"
         },
         "vocoder": {
-          "bytes": 38613392,
           "quantized": "dynamic_wi8_afp32"
         }
       },
@@ -623,7 +655,7 @@
       "exclude_context_phones": [
         "<sil>"
       ],
-      "hidden_size": 512,
       "layout": "balanced_before_target_after",
       "scope": "passage",
       "segments": 3,
@@ -726,7 +758,7 @@
   },
   "dataset_summary": {
     "controls": "measured_four_axes",
-    "voice_locale_pairs": 60
   },
   "default_language": "en_us",
   "languages": [
@@ -913,6 +945,74 @@
       "styles": [
         "default"
       ]
     }
   ]
 }

         0.0,
         4.0
       ],
+      "release_id": "v2-20261020",
       "scope": "utterance",
       "timeline_conditioning": false,
       "voice_locales": {
           "it",
           "vi"
         ],
+        "kincade": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "max": [
           "de",
           "en_us",
           "it",
           "vi"
         ],
+        "poppy": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
+        "priya": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "rex": [
           "de",
           "en_us",
           "it",
           "vi"
         ],
+        "thistle": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "tuesday": [
           "de",
           "en_gb",
       "quantization": "dynamic_wi8_afp32",
       "report": {
         "duration_predictor": {
+          "bytes": 11668736,
           "quantized": "dynamic_wi8_afp32"
         },
         "vector_context_encoder": {
+          "bytes": 11894656,
           "quantized": "dynamic_wi8_afp32"
         },
         "vector_estimator": {
+          "bytes": 146540928,
           "quantized": "dynamic_wi8_afp32"
         },
         "vocoder": {
+          "bytes": 38615216,
           "quantized": "dynamic_wi8_afp32"
         }
       },
       "exclude_context_phones": [
         "<sil>"
       ],
+      "hidden_size": 640,
       "layout": "balanced_before_target_after",
       "scope": "passage",
       "segments": 3,
   },
   "dataset_summary": {
     "controls": "measured_four_axes",
+    "voice_locale_pairs": 84
   },
   "default_language": "en_us",
   "languages": [
       "styles": [
         "default"
       ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "thistle",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "thistle",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "kincade",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "kincade",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "poppy",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "poppy",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "priya",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "priya",
+      "styles": [
+        "default"
+      ]
     }
   ]
 }

onnx/assets/fixed_references.json CHANGED Viewed

The diff for this file is too large to render. See raw diff

onnx/assets/voices.json CHANGED Viewed

@@ -38,5 +38,21 @@
   {
     "id": "tuesday",
     "index": 9
   }
 ]

   {
     "id": "tuesday",
     "index": 9
+  },
+  {
+    "id": "thistle",
+    "index": 10
+  },
+  {
+    "id": "kincade",
+    "index": 11
+  },
+  {
+    "id": "poppy",
+    "index": 12
+  },
+  {
+    "id": "priya",
+    "index": 13
   }
 ]

onnx/manifest.json CHANGED Viewed

@@ -432,7 +432,7 @@
         0.0,
         4.0
       ],
-      "release_id": "v2-20261006",
       "scope": "utterance",
       "timeline_conditioning": false,
       "voice_locales": {
@@ -468,6 +468,14 @@
           "it",
           "vi"
         ],
         "max": [
           "de",
           "en_us",
@@ -484,6 +492,22 @@
           "it",
           "vi"
         ],
         "rex": [
           "de",
           "en_us",
@@ -508,6 +532,14 @@
           "it",
           "vi"
         ],
         "tuesday": [
           "de",
           "en_gb",
@@ -546,7 +578,7 @@
       ],
       "shared_initializer_map": "onnx/components/shared_weights.json",
       "shared_weights": "onnx/components/shared_weights.bin",
-      "shared_weights_bytes": 208219968,
       "quantization": {
         "format": "onnx_dynamic_int8",
         "weight_type": "QInt8",
@@ -562,22 +594,22 @@
             "int8_nodes": 8
           },
           "vector_estimator": {
-            "int8_nodes": 72
           },
           "vector_estimator_128": {
-            "int8_nodes": 72
           },
           "vector_estimator_256": {
-            "int8_nodes": 72
           },
           "vector_estimator_384": {
-            "int8_nodes": 72
           },
           "vector_estimator_512": {
-            "int8_nodes": 72
           },
           "vector_estimator_768": {
-            "int8_nodes": 72
           },
           "vocoder": {
             "int8_nodes": 0
@@ -646,7 +678,7 @@
       "exclude_context_phones": [
         "<sil>"
       ],
-      "hidden_size": 512,
       "layout": "balanced_before_target_after",
       "scope": "passage",
       "segments": 3,
@@ -749,7 +781,7 @@
   },
   "dataset_summary": {
     "controls": "measured_four_axes",
-    "voice_locale_pairs": 60
   },
   "default_language": "en_us",
   "languages": [
@@ -936,6 +968,74 @@
       "styles": [
         "default"
       ]
     }
   ]
 }

         0.0,
         4.0
       ],
+      "release_id": "v2-20261020",
       "scope": "utterance",
       "timeline_conditioning": false,
       "voice_locales": {
           "it",
           "vi"
         ],
+        "kincade": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "max": [
           "de",
           "en_us",
           "it",
           "vi"
         ],
+        "poppy": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
+        "priya": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "rex": [
           "de",
           "en_us",
           "it",
           "vi"
         ],
+        "thistle": [
+          "en_us",
+          "es",
+          "de",
+          "fr",
+          "it",
+          "vi"
+        ],
         "tuesday": [
           "de",
           "en_gb",
       ],
       "shared_initializer_map": "onnx/components/shared_weights.json",
       "shared_weights": "onnx/components/shared_weights.bin",
+      "shared_weights_bytes": 280418752,
       "quantization": {
         "format": "onnx_dynamic_int8",
         "weight_type": "QInt8",
             "int8_nodes": 8
           },
           "vector_estimator": {
+            "int8_nodes": 96
           },
           "vector_estimator_128": {
+            "int8_nodes": 96
           },
           "vector_estimator_256": {
+            "int8_nodes": 96
           },
           "vector_estimator_384": {
+            "int8_nodes": 96
           },
           "vector_estimator_512": {
+            "int8_nodes": 96
           },
           "vector_estimator_768": {
+            "int8_nodes": 96
           },
           "vocoder": {
             "int8_nodes": 0
       "exclude_context_phones": [
         "<sil>"
       ],
+      "hidden_size": 640,
       "layout": "balanced_before_target_after",
       "scope": "passage",
       "segments": 3,
   },
   "dataset_summary": {
     "controls": "measured_four_axes",
+    "voice_locale_pairs": 84
   },
   "default_language": "en_us",
   "languages": [
       "styles": [
         "default"
       ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "thistle",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "thistle",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "kincade",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "kincade",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "poppy",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "poppy",
+      "styles": [
+        "default"
+      ]
+    },
+    {
+      "default_language": "en_us",
+      "embedding_path": null,
+      "id": "priya",
+      "languages": [
+        "en_us",
+        "es",
+        "de",
+        "fr",
+        "it",
+        "vi"
+      ],
+      "name": "priya",
+      "styles": [
+        "default"
+      ]
     }
   ]
 }