SenseVoiceSmall community API proposal: transcription-only task and language mapping
Proposing a community local-transcription example for SenseVoiceSmall via Rapid-MLX on Apple Silicon, using mlx-audio.
The adapter explicitly maps language hints to SenseVoice's supported set and falls back to auto detection for unknown codes. It rejects translation requests instead of silently treating them as transcription. The linked tests cover this task boundary and metadata-based backend detection.
The proposed example would use /v1/audio/transcriptions, a pinned checkpoint and an audio sample with a known transcript. It should not imply that the API exposes all of SenseVoice's emotion/event tags or supports Whisper's translation behavior. A real transcript and audio-format receipt remain required before labeling the commands qualified; the model-specific request contract is available now for community review.
Concrete catalog entry: sensevoice-small → mlx-community/SenseVoiceSmall@8ddd966bd96243cff196422f81f0c5d955814792. This is a conversion/distribution artifact; it is not presented as an official upstream release.
Source references (fixed Rapid commit): catalog, SenseVoice adapter, task/language tests, audio guide.
Verification status: this proposal is based on reading the catalog, implementation and the linked documentation/tests. No fresh model inference or test-suite run was performed for this submission; source tests are not an end-to-end qualification. It requests a community documentation scope, not an official endorsement or a performance listing.
Disclosure: I maintain Rapid-MLX. Prepared with AI assistance under my direction. Anonymous usage-metadata telemetry is enabled by default; DO_NOT_TRACK=1 disables it. The original model and conversion terms continue to apply.