absorb.md

Speech Recognition

No compiled wiki article for this topic yet. Raw entries below are the source material — a wiki article can be generated on demand from /admin/triggers.

WhisperPipe: Bounded-Memory Streaming ASR Achieves Near-Offline Accuracy at 3–5x Lower Latency

WhisperPipe introduces a streaming architecture for real-time ASR built on top of Whisper that resolves the classic accuracy-vs-efficiency tradeoff through three targeted innovations: a hybrid VAD pipeline, dynamic overlapping context buffers, and adaptive processing. On 2.5 hours of diverse audio,