Breeze ASR 25 for WhisperLiveKit
This repository repackages Breeze ASR 25 into a layout that works directly with WhisperLiveKit custom models and also remains usable with faster-whisper / CTranslate2.
It combines:
- the original Whisper decoder checkpoint from
MediaTek-Research/Breeze-ASR-25 - the CTranslate2 conversion from
SoybeanMilk/faster-whisper-Breeze-ASR-25
This is a compatibility packaging repo. It does not introduce new training or new weights.
Included files
breeze-asr-25.pt: Whisper decoder checkpoint for WhisperLiveKit SimulStreamingmodel.bin: CTranslate2 encoder/decoder weights for faster-whisperconfig.json,preprocessor_config.json,tokenizer.json,vocabulary.json: faster-whisper runtime files
WhisperLiveKit usage
Point WhisperLiveKit to this Hugging Face repo ID or to a local snapshot of this directory.
Example:
whisperlivekit-server \
--backend simulstreaming \
--model-path YOUR_USERNAME/YOUR_REPO_NAME
If you want to set alignment heads explicitly, use WhisperLiveKit's scripts/determine_alignment_heads.py against breeze-asr-25.pt and pass the result with --custom-alignment-heads.
faster-whisper usage
from faster_whisper import WhisperModel
model = WhisperModel("YOUR_USERNAME/YOUR_REPO_NAME", compute_type="float16")
segments, info = model.transcribe("audio.wav")
for segment in segments:
print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")
Provenance
- Original model:
MediaTek-Research/Breeze-ASR-25 - Existing CTranslate2 conversion:
SoybeanMilk/faster-whisper-Breeze-ASR-25 - Compatibility packaging assembled for WhisperLiveKit custom-model usage from those two upstream artifacts
Credits
Original Breeze ASR 25 authors
Please credit the Breeze ASR 25 authors and cite their paper:
- Cheng-Kang Chou
- Chan-Jan Hsu
- Ho-Lam Chung
- Liang-Hsuan Tseng
- Hsi-Chun Cheng
- Yu-Kuan Fu
- Kuan-Po Huang
- Hung-yi Lee
Paper: A Self-Refining Framework for Enhancing ASR Using TTS-Synthesized Data
Conversion credit
The faster-whisper / CTranslate2 conversion included here is based on SoybeanMilk/faster-whisper-Breeze-ASR-25 by SoybeanMilk.
Usage note
If you redistribute or publish derived packaging of this repo, keep the upstream attribution to both the original Breeze ASR 25 authors and the CTranslate2 conversion author.
- Downloads last month
- 37
Model tree for Vonernue/WhisperLiveKit-Breeze-ASR-25
Base model
openai/whisper-large-v2