Show HN: Audio AI had a wild day – 5 major open-source / real-time TTS drops

The audio/TTS space just moved fast. In the last week alone:

NVIDIA – PersonaPlex-7B Open-source, full-duplex conversational speech model.

Inworld AI – TTS-1.5 Realtime TTS (<250ms), $0.005/min, currently #1 on Artificial Analysis.

Flash Labs – Chroma 1.0 First open-source, end-to-end, real-time speech-to-speech model.

Alibaba Qwen – Qwen3-TTS Fully open-sourced TTS family: Base, CustomVoice, VoiceDesign.

Kyutai Labs – Pocket TTS Runs locally on a laptop. No GPU required.

Feels like TTS is hitting the same acceleration moment LLMs had last year. Realtime, open-source, and local is becoming the default.

Curious what people here are building with this