Add Phonon-2, Underdog Bark 3, and Haas ASR packages

.gitattributes

@@ -255,3 +255,6 @@ KittenTTS2-GGUF/kitten-tts2-20261021-q8_0.gguf filter=lfs diff=lfs merge=lfs -te
 KittenTTS2-GGUF/kitten-tts2-20261021-tq2_1.gguf filter=lfs diff=lfs merge=lfs -text
 Irodori-TTS-v4-Small-GGUF/irodori-tts-v4.1-small-mf-f16.gguf filter=lfs diff=lfs merge=lfs -text
 Irodori-TTS-v4-Small-GGUF/irodori-tts-v4.1-small-mf-q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+Haas-ja-v1.0-GGUF/haas-ja-v1.0-f16.gguf filter=lfs diff=lfs merge=lfs -text
+Haas-ja-v1.0-GGUF/haas-ja-v1.0-q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+Underdog-Bark-3.0-GGUF/underdog-bark-3.0-mixed-f16.gguf filter=lfs diff=lfs merge=lfs -text

Haas-ja-v1.0-GGUF/README.md

@@ -0,0 +1,52 @@
+---
+library_name: audio.cpp
+license: apache-2.0
+language:
+- ja
+tags:
+- audio.cpp
+- gguf
+- automatic-speech-recognition
+base_model:
+- third-intelligence/Haas-ja-v1.0
+---
+
+# Haas-ja-v1.0 GGUF
+
+Follow [audio.cpp](https://github.com/0xShug0/audio.cpp) for runtime support.
+Converted from [third-intelligence/Haas-ja-v1.0](https://huggingface.co/third-intelligence/Haas-ja-v1.0/tree/660728d39019e5e065046154beca44ac9a66a540).
+
+Japanese FastConformer-TDT ASR with a Kimi Delta Attention predictor. Use the
+`haas_asr` family with the F16 or Q8_0 GGUF.
+
+```bash
+audiocpp_cli --task asr --family haas_asr \
+  --model haas-ja-v1.0-q8_0.gguf --backend cuda \
+  --session-option haas_asr.matmul_weight_type=q8_0 \
+  --audio japanese.wav --text-out transcript.txt --log
+```
+
+The encoder is full-context. Buffered streaming uses audio.cpp's Parakeet
+interface; this is not a native cache-aware streaming checkpoint.
+
+The existing `haas_asr.matmul_weight_type=q8_0` session option also keeps
+load-time packed and scaled projections in Q8_0. With `native`, projections
+constructed from float32 values retain float32 storage. Omit the override when
+using the F16 package if that precision is preferred.
+
+## CUDA performance
+
+RTX 5090, Debug audio.cpp build, one session with a short warmup followed by
+24 Japanese JSUT test utterances (97.91 seconds total), one utterance per
+request. audio.cpp uses Q8_0 weights and
+`haas_asr.matmul_weight_type=q8_0`; official Haas Python uses its default
+BF16 precision. Warmup is excluded from RTF. Peak VRAM is the observed process
+peak across model loading and inference, sampled every 10 ms.
+
+| Runtime | RTF ↓ | Peak VRAM |
+| --- | ---: | ---: |
+| Official Python | 0.00690 | 4,326 MiB |
+| audio.cpp Q8_0 | 0.00379 | 1,335 MiB |
+
+This workload was **1.82× faster** with **69% less peak VRAM** in audio.cpp.
+These are a small workload's measurements, not a full-corpus benchmark.

Haas-ja-v1.0-GGUF/haas-ja-v1.0-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b283c8ca3fcf50d06026f7caa309f5f7cddd0e6fc9fbc5889eb1b66f8a310edc
+size 1232193568

Haas-ja-v1.0-GGUF/haas-ja-v1.0-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9112e512a93067161440e4f096066a739f885f13b9b271ebdafb70438a17ef0
+size 880730576

Phonon-2-GGUF/phonon-2-mixed-f16.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0455e006b69c9878f0c551aeb565f6fc892c054419f3c231ff1153b7979f4cf7
-size 1301738720
+oid sha256:de6e2e26ce1e924f1bf4b0c9c0f3e9414f8e7ce5597df4554158c4a211e8aacb
+size 1301737856

Phonon-2-GGUF/phonon-2-mixed-q8.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47dfe1ed5d403e3d8e97fdb39489c233df8ea25a408eab3a19ea251566b4ab3b
-size 1183641824
+oid sha256:8245772967def6edbb7ac54d2c453ecc13040687a5709c83d4200af339b51ff2
+size 1183640960

README.md

@@ -26,8 +26,10 @@ base_model:
 - Aratako/MioTTS-1.7B
 - Aratako/Semantic-DACVAE-Japanese-32dim
 - Banafo/Kroko-ASR
+- ConwayResearch/Underdog-Bark-3.0
 - dots-studio/dots.tts-mf
 - dots-studio/dots.tts-soar
+- FermionResearch/Phonon-2
 - fishaudio/s2-pro
 - FunAudioLLM/Fun-ASR-Nano-2512-hf
 - ai-sage/GigaAM-Multilingual
@@ -79,6 +81,7 @@ base_model:
 - stabilityai/stable-audio-3-medium
 - stabilityai/stable-audio-3-small-music
 - stabilityai/stable-audio-3-small-sfx
+- third-intelligence/Haas-ja-v1.0
 ---
 
 # audio.cpp GGUF Model Packages
@@ -129,6 +132,7 @@ names follow audio.cpp's [model license reference](https://github.com/0xShug0/au
 | `Fun-ASR-Nano-2512-GGUF` | `fun_asr_nano` | F16 + Q8 | Apache-2.0 |
 | `GigaAM-ASR-GGUF` | `gigaam_asr` | F16 + F32 | MIT |
 | `Granite-Speech-5.0-470M-TurboCTC-GGUF` | `granite5asr` | Q8 | Apache-2.0 |
+| `Haas-ja-v1.0-GGUF` | `haas_asr` | F16 + Q8 | Apache-2.0 |
 | `HeartMuLa-GGUF` | `heartmula` | F16 + Q8 | Apache-2.0 |
 | `HTDemucs-GGUF` | `htdemucs` | F16 + Q8 | MIT |
 | `HTDemucs-6stems-GGUF` | `htdemucs_6stems` | F16 + Q8 | MIT |
@@ -173,7 +177,7 @@ names follow audio.cpp's [model license reference](https://github.com/0xShug0/au
 | `OpenWakeWord-GGUF` | `open_wake_word` | F32 | CC-BY-NC-SA-4.0 |
 | `Parakeet-TDT-0.6B-v3-GGUF` | `parakeet_tdt` | F16 + Q8 | CC-BY-4.0 |
 | `PersonaPlex-GGUF` | `personaplex` | Q4_K + Q8 | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) |
-| `Phonon-2-GGUF` | `parakeet_tdt` | mixed F16 + Q8 | CC-BY-4.0 |
+| `Phonon-2-GGUF` | `phonon2_asr` | mixed F16 + mixed Q8 | CC-BY-4.0 |
 | `Piper-TTS-GGUF` | `piper_tts` | original | MIT |
 | `Plapre-3-GGUF` | `plapre` | F32 + BF16 | CC-BY-4.0 |
 | `PocketTTS-GGUF` | `pocket_tts` | BF16 + Q8 | CC-BY-4.0 |
@@ -198,6 +202,7 @@ names follow audio.cpp's [model license reference](https://github.com/0xShug0/au
 | `Supertonic-3-GGUF` | `supertonic` | original + F16 + Q8 | [BigScience OpenRAIL-M](https://huggingface.co/Supertone/supertonic-3/blob/main/LICENSE) |
 | `TF-GridNet-GGUF` | `tf_gridnet` | F32 | Not stated |
 | `Tone-Color-VC-GGUF` | `tone_color_vc` | F32 + F16 | MIT |
+| `Underdog-Bark-3.0-GGUF` | `underdog_bark3` | mixed F16 (Standard) | CC-BY-4.0 |
 | `UniverSR-GGUF` | `universr` | original | CC-BY-4.0 |
 | `Vevo2-GGUF` | `vevo2` | original + F16 + Q8 | CC-BY-NC-ND-4.0 |
 | `VibeVoice-1.5B-GGUF` | `vibevoice` | BF16 + Q8 + Q4 | MIT |

Underdog-Bark-3.0-GGUF/underdog-bark-3.0-mixed-f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bd42328378d0ff25d8ef3ac22a2d7e83deffe6a2481afc1a240218d046671d1
+size 1301642208