speech-to-speech · Issues· 114 open

Listener backchannels abort active response mid-playback

STT can hallucinate confident, mislabeled transcriptions with no confidence signal, leading to nonsensical input reaching the LLM

Preserve Responses API reasoning items during tool continuations

No user-facing signal while a response is generating on slower local TTS/LLM backends

Sending silence on RTC causes noticable per-client background CPU usage

Browser realtime demo can underrun TTS playback without a startup jitter buffer

Add FunASR/SenseVoice as a self-hosted STT option

STT handlers retain the detected language across sessions; only Parakeet resets it

Speculative execution during user speech?

Support optional mid-session model switching through session.update

Feature Request: Great new VAD and ASR models

llama-swap support

Maintainer on holidays

Reduce barge-in latency with reversible pre-confirmation audio ducking

Support stateful streaming STT sessions instead of repeated whole-utterance uploads