A turn-taking model that knows when a voice agent should speak
Emerged finds new products worth a look, daily — what’s new today
audio-chat-box
A local side process that gives your app a voice and ears via JSON lines
Is this yours? Claim itIf you want an app or agent to talk and listen without a cloud speech API, this puts Kokoro and Parakeet behind a pipe and keeps GPL code out of your binary.
audio-chat-box runs as its own process. Your program writes a JSON line to speak text with Kokoro or to transcribe audio with NVIDIA's Parakeet, and reads the answer back; Silero detects when someone starts and stops talking. Your app still owns the mic and speakers. Models are about 835 MB, fetched once and hash-checked. English only, one listener at a time, Rust to build. A Docker demo offers 54 voices. By Carlos Justiniano, from his chess coach.
More like this
A level-tracking check on top of Silero that ends turns sooner
Voice agents that run entirely on CPUs, in the cloud, on-prem or on-device
Free on-device dictation and read-aloud for Apple silicon Macs
A voice assistant that brainstorms with you, then runs and watches Claude Code
Hold a key, say what you need, and a sandboxed worker on your Mac does it