SenseVoice-Small
Open multilingual speech understanding — ASR plus emotion and audio events, fast
About
SenseVoice-Small is FunAudioLLM's open speech understanding model, open-sourced in July 2024. It is a non-autoregressive encoder that jointly handles speech recognition, language identification, emotion recognition, and audio event detection across Mandarin, Cantonese, English, Japanese, and Korean (community forks extend coverage toward 50+ languages). Community reports put it around 15x faster than Whisper at comparable accuracy. Deployment is flexible: ONNX and libtorch exports, and a llama.cpp/GGUF port that runs on CPU as a single ~254MB binary with built-in VAD — genuinely edge-friendly. It composes with FunASR's VAD, speaker, and punctuation models for diarized output. Limits: emotion and event detection are coarse; diarization is not native to the checkpoint; and documentation skews toward Chinese-language sources.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Community-reported: ~15x faster than Whisper at comparable accuracy
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: FunAudioLLM open-sourced July 2024; non-autoregressive encoder, 5 core languages; Community-reported ~15x faster than Whisper; GGUF/llama.cpp CPU port verified in READMEs; Open weights, reproducible.