Whisper-based voice bots are hitting 1.5-second lag and breaking on interruptions
Dull_Look_2320 · reddit · 2026-07-27
Building a conversational voice bot with the standard OpenAI Whisper API is creating a 1.5-second end-to-end delay once STT, LLM, and TTS are chained together.
The author also says the stack breaks when users interrupt or talk over the bot, because the STT layer cannot handle live crosstalk or diarization well. They ask for production-ready low-latency setups and for ways to avoid months of custom VAD and chunking work.
More from Infra
- AI Shipping Labs opens code for 10 workshops on agents, vLLM, and vector search — Al_Grigor · 2026-07-27
- Observer builds local agents that watch your screen, mic, and camera for alerts — alex_verem · 2026-07-27
- What OS and coding models should power a dedicated local AI box with a 3090 and 5060 Ti? — PolarNightProphecies · 2026-07-27
- JAX-JS makes Kyutai Pocket TTS run at 2.5× realtime in the browser on Wasm CPU — dejavucoder · 2026-07-27
- Moonshot says Kimi-K3 open weights are due today, but it needs 64+ accelerators — heypearlai · 2026-07-27
- llmux manages vLLM and llama.cpp model swaps with one profile per model — Available-Message509 · 2026-07-27