#speech-to-speech

Mistral said on April 2, 2026 that developers can assemble a web-search-enabled speech-to-speech assistant in roughly 150 lines of code using Voxtral for transcription and speech generation plus Mistral Small 4 for agentic reasoning. The post is notable less as a single model launch than as a clear reference architecture for real-time audio agents.

An HN post on a Swift/MLX port of Nvidia PersonaPlex 7B shows how chunking, buffering, and interrupt handling matter as much as raw model quality for local speech-to-speech agents.