Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
AudioStream -> List<AudioChunk>
Segment an audio stream into batched chunks based on voice activity boundaries to enable parallel ASR processing and reduce silence-induced hallucinations.
Problem it solves
Sequential transcription is slow, and long silent stretches trigger hallucinations in autoregressive ASR models.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.