voice + int8 onnx weights for the in-browser storyteller

README.md

@@ -0,0 +1,65 @@
+---
+license: mit
+language:
+  - en
+tags:
+  - text-to-speech
+  - onnx
+  - browser
+  - webassembly
+library_name: onnx
+---
+
+# pocket-tts-voice
+
+Weights for a text-to-speech demo that runs entirely in a browser tab, with no
+server: https://gaurav.bar/things/storyteller
+
+Two things live here.
+
+## 1. A voice
+
+`onnx/english_2026-04/voices.bin` is **my own voice** — Gaurav Singh — encoded
+once, offline, from about twenty seconds of me speaking. I am publishing it
+knowingly, so that the demo works for anyone who opens the page.
+
+The file is not audio. It is the speech model's key–value cache after it has
+listened to that recording: 12,288 floats for every 80 ms, across six layers and
+sixteen attention heads. Encoding offline means the browser never needs the
+voice encoder (19.8 MB) or the gated cloning weights — it loads a state file, as
+the upstream built-in voices do.
+
+**If you use this voice, please say that it is synthetic.** The demo page it was
+built for discloses that in its first sentence.
+
+## 2. Redistributed model weights
+
+`flow_lm_*`, `mimi_decoder`, `text_conditioner`, `tokenizer.model` and
+`bos_before_voice.npy` are the int8 ONNX exports of
+[Pocket TTS](https://github.com/kyutai-labs/pocket-tts) (Kyutai, MIT),
+mirrored from
+[KevinAHM/pocket-tts-onnx](https://huggingface.co/KevinAHM/pocket-tts-onnx)
+so one origin serves the whole demo.
+
+`mimi_encoder` is deliberately **not** mirrored — it is only needed to encode a
+new voice, which this demo does not do.
+
+### Local change
+
+`bundle.json` lists `predefined_voices: ["gaurav"]` rather than the upstream
+eight. The runtime defaults to `alba` whenever it sees that name in the list and
+then fails against a pack that does not contain it.
+
+## Layout
+
+```
+onnx/english_2026-04/
+  flow_lm_main_int8.onnx        72.8 MB
+  mimi_decoder_int8.onnx        21.6 MB
+  text_conditioner_int8.onnx    15.6 MB
+  flow_lm_flow_int8.onnx         9.5 MB
+  voices.bin                    11.8 MB   <- the voice
+  bundle.json, tokenizer.model, bos_before_voice.npy
+```
+
+About 133 MB in total, fetched once and then cached by the browser.

onnx/english_2026-04/bos_before_voice.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f46edf4f7007b7ba4ea58831f49d003e59e167b4641c44bb3addfe9231a780b1
+size 4224

onnx/english_2026-04/bundle.json

@@ -0,0 +1,1074 @@
+{
+  "bos_before_voice_file": "bos_before_voice.npy",
+  "bundle_name": "english_2026-04",
+  "conditioning_dim": 1024,
+  "flow_lm_state_manifest": [
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 0,
+      "input_name": "state_0",
+      "key": "cache",
+      "module": "transformer.layers.0.self_attn",
+      "output_name": "out_state_0",
+      "path": "transformer.layers.0.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 1,
+      "input_name": "state_1",
+      "key": "current_end",
+      "module": "transformer.layers.0.self_attn",
+      "output_name": "out_state_1",
+      "path": "transformer.layers.0.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 2,
+      "input_name": "state_2",
+      "key": "step",
+      "module": "transformer.layers.0.self_attn",
+      "output_name": "out_state_2",
+      "path": "transformer.layers.0.self_attn/step",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 3,
+      "input_name": "state_3",
+      "key": "cache",
+      "module": "transformer.layers.1.self_attn",
+      "output_name": "out_state_3",
+      "path": "transformer.layers.1.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 4,
+      "input_name": "state_4",
+      "key": "current_end",
+      "module": "transformer.layers.1.self_attn",
+      "output_name": "out_state_4",
+      "path": "transformer.layers.1.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 5,
+      "input_name": "state_5",
+      "key": "step",
+      "module": "transformer.layers.1.self_attn",
+      "output_name": "out_state_5",
+      "path": "transformer.layers.1.self_attn/step",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 6,
+      "input_name": "state_6",
+      "key": "cache",
+      "module": "transformer.layers.2.self_attn",
+      "output_name": "out_state_6",
+      "path": "transformer.layers.2.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 7,
+      "input_name": "state_7",
+      "key": "current_end",
+      "module": "transformer.layers.2.self_attn",
+      "output_name": "out_state_7",
+      "path": "transformer.layers.2.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 8,
+      "input_name": "state_8",
+      "key": "step",
+      "module": "transformer.layers.2.self_attn",
+      "output_name": "out_state_8",
+      "path": "transformer.layers.2.self_attn/step",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 9,
+      "input_name": "state_9",
+      "key": "cache",
+      "module": "transformer.layers.3.self_attn",
+      "output_name": "out_state_9",
+      "path": "transformer.layers.3.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 10,
+      "input_name": "state_10",
+      "key": "current_end",
+      "module": "transformer.layers.3.self_attn",
+      "output_name": "out_state_10",
+      "path": "transformer.layers.3.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 11,
+      "input_name": "state_11",
+      "key": "step",
+      "module": "transformer.layers.3.self_attn",
+      "output_name": "out_state_11",
+      "path": "transformer.layers.3.self_attn/step",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 12,
+      "input_name": "state_12",
+      "key": "cache",
+      "module": "transformer.layers.4.self_attn",
+      "output_name": "out_state_12",
+      "path": "transformer.layers.4.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 13,
+      "input_name": "state_13",
+      "key": "current_end",
+      "module": "transformer.layers.4.self_attn",
+      "output_name": "out_state_13",
+      "path": "transformer.layers.4.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 14,
+      "input_name": "state_14",
+      "key": "step",
+      "module": "transformer.layers.4.self_attn",
+      "output_name": "out_state_14",
+      "path": "transformer.layers.4.self_attn/step",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "nan",
+      "index": 15,
+      "input_name": "state_15",
+      "key": "cache",
+      "module": "transformer.layers.5.self_attn",
+      "output_name": "out_state_15",
+      "path": "transformer.layers.5.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        1000,
+        16,
+        64
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 16,
+      "input_name": "state_16",
+      "key": "current_end",
+      "module": "transformer.layers.5.self_attn",
+      "output_name": "out_state_16",
+      "path": "transformer.layers.5.self_attn/current_end",
+      "shape": [
+        0
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 17,
+      "input_name": "state_17",
+      "key": "step",
+      "module": "transformer.layers.5.self_attn",
+      "output_name": "out_state_17",
+      "path": "transformer.layers.5.self_attn/step",
+      "shape": [
+        1
+      ]
+    }
+  ],
+  "frame_rate": 12.5,
+  "insert_bos_before_voice": true,
+  "language": "english_2026-04",
+  "latent_dim": 32,
+  "max_token_per_chunk": 50,
+  "mimi_state_manifest": [
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 0,
+      "input_name": "state_0",
+      "key": "first",
+      "module": "decoder.model.0",
+      "output_name": "out_state_0",
+      "path": "decoder.model.0/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 1,
+      "input_name": "state_1",
+      "key": "previous",
+      "module": "decoder.model.0",
+      "output_name": "out_state_1",
+      "path": "decoder.model.0/previous",
+      "shape": [
+        1,
+        512,
+        6
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 2,
+      "input_name": "state_2",
+      "key": "first",
+      "module": "decoder.model.11",
+      "output_name": "out_state_2",
+      "path": "decoder.model.11/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 3,
+      "input_name": "state_3",
+      "key": "previous",
+      "module": "decoder.model.11",
+      "output_name": "out_state_3",
+      "path": "decoder.model.11/previous",
+      "shape": [
+        1,
+        64,
+        2
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 4,
+      "input_name": "state_4",
+      "key": "partial",
+      "module": "decoder.model.2",
+      "output_name": "out_state_4",
+      "path": "decoder.model.2/partial",
+      "shape": [
+        1,
+        256,
+        6
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 5,
+      "input_name": "state_5",
+      "key": "first",
+      "module": "decoder.model.3.block.1",
+      "output_name": "out_state_5",
+      "path": "decoder.model.3.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 6,
+      "input_name": "state_6",
+      "key": "previous",
+      "module": "decoder.model.3.block.1",
+      "output_name": "out_state_6",
+      "path": "decoder.model.3.block.1/previous",
+      "shape": [
+        1,
+        256,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 7,
+      "input_name": "state_7",
+      "key": "first",
+      "module": "decoder.model.3.block.3",
+      "output_name": "out_state_7",
+      "path": "decoder.model.3.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 8,
+      "input_name": "state_8",
+      "key": "previous",
+      "module": "decoder.model.3.block.3",
+      "output_name": "out_state_8",
+      "path": "decoder.model.3.block.3/previous",
+      "shape": [
+        1,
+        128,
+        0
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 9,
+      "input_name": "state_9",
+      "key": "partial",
+      "module": "decoder.model.5",
+      "output_name": "out_state_9",
+      "path": "decoder.model.5/partial",
+      "shape": [
+        1,
+        128,
+        5
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 10,
+      "input_name": "state_10",
+      "key": "first",
+      "module": "decoder.model.6.block.1",
+      "output_name": "out_state_10",
+      "path": "decoder.model.6.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 11,
+      "input_name": "state_11",
+      "key": "previous",
+      "module": "decoder.model.6.block.1",
+      "output_name": "out_state_11",
+      "path": "decoder.model.6.block.1/previous",
+      "shape": [
+        1,
+        128,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 12,
+      "input_name": "state_12",
+      "key": "first",
+      "module": "decoder.model.6.block.3",
+      "output_name": "out_state_12",
+      "path": "decoder.model.6.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 13,
+      "input_name": "state_13",
+      "key": "previous",
+      "module": "decoder.model.6.block.3",
+      "output_name": "out_state_13",
+      "path": "decoder.model.6.block.3/previous",
+      "shape": [
+        1,
+        64,
+        0
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 14,
+      "input_name": "state_14",
+      "key": "partial",
+      "module": "decoder.model.8",
+      "output_name": "out_state_14",
+      "path": "decoder.model.8/partial",
+      "shape": [
+        1,
+        64,
+        4
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 15,
+      "input_name": "state_15",
+      "key": "first",
+      "module": "decoder.model.9.block.1",
+      "output_name": "out_state_15",
+      "path": "decoder.model.9.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 16,
+      "input_name": "state_16",
+      "key": "previous",
+      "module": "decoder.model.9.block.1",
+      "output_name": "out_state_16",
+      "path": "decoder.model.9.block.1/previous",
+      "shape": [
+        1,
+        64,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 17,
+      "input_name": "state_17",
+      "key": "first",
+      "module": "decoder.model.9.block.3",
+      "output_name": "out_state_17",
+      "path": "decoder.model.9.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 18,
+      "input_name": "state_18",
+      "key": "previous",
+      "module": "decoder.model.9.block.3",
+      "output_name": "out_state_18",
+      "path": "decoder.model.9.block.3/previous",
+      "shape": [
+        1,
+        32,
+        0
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 19,
+      "input_name": "state_19",
+      "key": "cache",
+      "module": "decoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_19",
+      "path": "decoder_transformer.transformer.layers.0.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        8,
+        1000,
+        64
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 20,
+      "input_name": "state_20",
+      "key": "end_offset",
+      "module": "decoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_20",
+      "path": "decoder_transformer.transformer.layers.0.self_attn/end_offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 21,
+      "input_name": "state_21",
+      "key": "offset",
+      "module": "decoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_21",
+      "path": "decoder_transformer.transformer.layers.0.self_attn/offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 22,
+      "input_name": "state_22",
+      "key": "cache",
+      "module": "decoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_22",
+      "path": "decoder_transformer.transformer.layers.1.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        8,
+        1000,
+        64
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 23,
+      "input_name": "state_23",
+      "key": "end_offset",
+      "module": "decoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_23",
+      "path": "decoder_transformer.transformer.layers.1.self_attn/end_offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 24,
+      "input_name": "state_24",
+      "key": "offset",
+      "module": "decoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_24",
+      "path": "decoder_transformer.transformer.layers.1.self_attn/offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 25,
+      "input_name": "state_25",
+      "key": "first",
+      "module": "downsample.conv",
+      "output_name": "out_state_25",
+      "path": "downsample.conv/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 26,
+      "input_name": "state_26",
+      "key": "previous",
+      "module": "downsample.conv",
+      "output_name": "out_state_26",
+      "path": "downsample.conv/previous",
+      "shape": [
+        1,
+        512,
+        16
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 27,
+      "input_name": "state_27",
+      "key": "first",
+      "module": "encoder.model.0",
+      "output_name": "out_state_27",
+      "path": "encoder.model.0/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 28,
+      "input_name": "state_28",
+      "key": "previous",
+      "module": "encoder.model.0",
+      "output_name": "out_state_28",
+      "path": "encoder.model.0/previous",
+      "shape": [
+        1,
+        1,
+        6
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 29,
+      "input_name": "state_29",
+      "key": "first",
+      "module": "encoder.model.1.block.1",
+      "output_name": "out_state_29",
+      "path": "encoder.model.1.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 30,
+      "input_name": "state_30",
+      "key": "previous",
+      "module": "encoder.model.1.block.1",
+      "output_name": "out_state_30",
+      "path": "encoder.model.1.block.1/previous",
+      "shape": [
+        1,
+        64,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 31,
+      "input_name": "state_31",
+      "key": "first",
+      "module": "encoder.model.1.block.3",
+      "output_name": "out_state_31",
+      "path": "encoder.model.1.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 32,
+      "input_name": "state_32",
+      "key": "previous",
+      "module": "encoder.model.1.block.3",
+      "output_name": "out_state_32",
+      "path": "encoder.model.1.block.3/previous",
+      "shape": [
+        1,
+        32,
+        0
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 33,
+      "input_name": "state_33",
+      "key": "first",
+      "module": "encoder.model.11",
+      "output_name": "out_state_33",
+      "path": "encoder.model.11/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 34,
+      "input_name": "state_34",
+      "key": "previous",
+      "module": "encoder.model.11",
+      "output_name": "out_state_34",
+      "path": "encoder.model.11/previous",
+      "shape": [
+        1,
+        512,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 35,
+      "input_name": "state_35",
+      "key": "first",
+      "module": "encoder.model.3",
+      "output_name": "out_state_35",
+      "path": "encoder.model.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 36,
+      "input_name": "state_36",
+      "key": "previous",
+      "module": "encoder.model.3",
+      "output_name": "out_state_36",
+      "path": "encoder.model.3/previous",
+      "shape": [
+        1,
+        64,
+        4
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 37,
+      "input_name": "state_37",
+      "key": "first",
+      "module": "encoder.model.4.block.1",
+      "output_name": "out_state_37",
+      "path": "encoder.model.4.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 38,
+      "input_name": "state_38",
+      "key": "previous",
+      "module": "encoder.model.4.block.1",
+      "output_name": "out_state_38",
+      "path": "encoder.model.4.block.1/previous",
+      "shape": [
+        1,
+        128,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 39,
+      "input_name": "state_39",
+      "key": "first",
+      "module": "encoder.model.4.block.3",
+      "output_name": "out_state_39",
+      "path": "encoder.model.4.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 40,
+      "input_name": "state_40",
+      "key": "previous",
+      "module": "encoder.model.4.block.3",
+      "output_name": "out_state_40",
+      "path": "encoder.model.4.block.3/previous",
+      "shape": [
+        1,
+        64,
+        0
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 41,
+      "input_name": "state_41",
+      "key": "first",
+      "module": "encoder.model.6",
+      "output_name": "out_state_41",
+      "path": "encoder.model.6/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 42,
+      "input_name": "state_42",
+      "key": "previous",
+      "module": "encoder.model.6",
+      "output_name": "out_state_42",
+      "path": "encoder.model.6/previous",
+      "shape": [
+        1,
+        128,
+        5
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 43,
+      "input_name": "state_43",
+      "key": "first",
+      "module": "encoder.model.7.block.1",
+      "output_name": "out_state_43",
+      "path": "encoder.model.7.block.1/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 44,
+      "input_name": "state_44",
+      "key": "previous",
+      "module": "encoder.model.7.block.1",
+      "output_name": "out_state_44",
+      "path": "encoder.model.7.block.1/previous",
+      "shape": [
+        1,
+        256,
+        2
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 45,
+      "input_name": "state_45",
+      "key": "first",
+      "module": "encoder.model.7.block.3",
+      "output_name": "out_state_45",
+      "path": "encoder.model.7.block.3/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "empty",
+      "index": 46,
+      "input_name": "state_46",
+      "key": "previous",
+      "module": "encoder.model.7.block.3",
+      "output_name": "out_state_46",
+      "path": "encoder.model.7.block.3/previous",
+      "shape": [
+        1,
+        128,
+        0
+      ]
+    },
+    {
+      "dtype": "bool",
+      "fill": "ones",
+      "index": 47,
+      "input_name": "state_47",
+      "key": "first",
+      "module": "encoder.model.9",
+      "output_name": "out_state_47",
+      "path": "encoder.model.9/first",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 48,
+      "input_name": "state_48",
+      "key": "previous",
+      "module": "encoder.model.9",
+      "output_name": "out_state_48",
+      "path": "encoder.model.9/previous",
+      "shape": [
+        1,
+        256,
+        6
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 49,
+      "input_name": "state_49",
+      "key": "cache",
+      "module": "encoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_49",
+      "path": "encoder_transformer.transformer.layers.0.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        8,
+        1000,
+        64
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 50,
+      "input_name": "state_50",
+      "key": "end_offset",
+      "module": "encoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_50",
+      "path": "encoder_transformer.transformer.layers.0.self_attn/end_offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 51,
+      "input_name": "state_51",
+      "key": "offset",
+      "module": "encoder_transformer.transformer.layers.0.self_attn",
+      "output_name": "out_state_51",
+      "path": "encoder_transformer.transformer.layers.0.self_attn/offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 52,
+      "input_name": "state_52",
+      "key": "cache",
+      "module": "encoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_52",
+      "path": "encoder_transformer.transformer.layers.1.self_attn/cache",
+      "shape": [
+        2,
+        1,
+        8,
+        1000,
+        64
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 53,
+      "input_name": "state_53",
+      "key": "end_offset",
+      "module": "encoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_53",
+      "path": "encoder_transformer.transformer.layers.1.self_attn/end_offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "int64",
+      "fill": "zeros",
+      "index": 54,
+      "input_name": "state_54",
+      "key": "offset",
+      "module": "encoder_transformer.transformer.layers.1.self_attn",
+      "output_name": "out_state_54",
+      "path": "encoder_transformer.transformer.layers.1.self_attn/offset",
+      "shape": [
+        1
+      ]
+    },
+    {
+      "dtype": "float32",
+      "fill": "zeros",
+      "index": 55,
+      "input_name": "state_55",
+      "key": "partial",
+      "module": "upsample.convtr",
+      "output_name": "out_state_55",
+      "path": "upsample.convtr/partial",
+      "shape": [
+        1,
+        512,
+        16
+      ]
+    }
+  ],
+  "model_recommended_frames_after_eos": null,
+  "pad_with_spaces_for_short_inputs": false,
+  "predefined_voices": [
+    "gaurav"
+  ],
+  "remove_semicolons": false,
+  "sample_rate": 24000,
+  "samples_per_frame": 1920,
+  "schema_version": 2,
+  "tokenizer_file": "tokenizer.model"
+}

onnx/english_2026-04/flow_lm_flow_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dd781ee5abee9e195320bf0106bebd6372a852b3b36352524ee78b40554635d
+size 9962530

onnx/english_2026-04/flow_lm_main_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9bd8106b79a0192c1c43399ab938fb24900a95c1c599870d75a884e99000116
+size 76341079

onnx/english_2026-04/mimi_decoder_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3630450a3297a101792a6ac66619ebc70ab916b265e6220c2afaef8b1673f925
+size 22684077

onnx/english_2026-04/text_conditioner_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24c0642bff11a4e1435c004d0c2aa6c68e093fadfa35f0f0dba849ac67f25ac3
+size 16388384

onnx/english_2026-04/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d461765ae179566678c93091c5fa6f2984c31bbe990bf1aa62d92c64d91bc3f6
+size 59339

onnx/english_2026-04/voices.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cbd37676d63f0c1ba80ead7b81711e1a9a516f0b85b2ff4c957b4864396801b
+size 12337885