Feature Request
Haystack is excellent for building NLP/RAG pipelines. A Speech-to-Text component powered by FunASR would enable audio document processing in Haystack pipelines.
Use case: Audio/video files → FunASR transcription → text preprocessing → indexing → retrieval
Why FunASR?
- OpenAI-compatible API:
/v1/audio/transcriptions endpoint — easy to wrap as a Haystack component
- SenseVoice: Ultra-fast ASR (234M params), 50+ languages, 5-10x faster than Whisper
- Complete pipeline: VAD + ASR + punctuation + speaker diarization + timestamps
- Self-hosted: No API key, runs locally
Potential Haystack component:
from funasr import AutoModel
class FunASRTranscriber:
def __init__(self):
self.model = AutoModel(
model="paraformer-zh",
vad_model="fsmn-vad",
punc_model="ct-punc",
spk_model="cam++",
)
def run(self, audio_path: str):
result = self.model.generate(input=audio_path)
return {"documents": [Document(content=r["text"]) for r in result]}Tasks
Feature Request
Haystack is excellent for building NLP/RAG pipelines. A Speech-to-Text component powered by FunASR would enable audio document processing in Haystack pipelines.
Use case: Audio/video files → FunASR transcription → text preprocessing → indexing → retrieval
Why FunASR?
/v1/audio/transcriptionsendpoint — easy to wrap as a Haystack componentPotential Haystack component:
Tasks
mainbranchintegration:<your integration name>has been added to the list of labels for this repository