PERSONAPLEX: VOICE AND ROLE CONTROL FOR FULL DUPLEX CONVERSATIONAL SPEECH MODELS

Summary from the TLDR AI newsletter

Personaplex is a real-time speech-to-speech conversational model that jointly performs streaming speech understanding and speech generation. It operates on continuous audio and can predict both text tokens and audio tokens auto-regressively to produce its spoken responses. Personaplex's design allows the model to update its internal state based on users' ongoing speech while still producing fluent output audio, so it can support highly interactive conversations. The model is ready for commercial use.