Transcribe audio into LlamaIndex Documents with FunASR — self-hosted speech-to-text powered by SenseVoice / Paraformer / Fun-ASR-Nano. Runs locally, no cloud API; strong on Chinese and 50+ languages.
pip install llama-index-readers-funasrfrom llama_index.readers.funasr import FunASRReader
reader = FunASRReader(model="iic/SenseVoiceSmall", device="cuda")
documents = reader.load_data("meeting.wav")
print(documents[0].text)Pass hub="hf" with model="FunAudioLLM/SenseVoiceSmall" to load from HuggingFace. The built-in VAD handles audio of any length.