Show HN: Hlas-1: open-weight 3B params TTS with frontier performance

Category: ai-ml

Tags: text-to-speech, tts, voice-cloning, streaming, pytorch, multilingual

Score: 6.8/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 7)

Breeze TTS 2 (Hlas-1) is an open-weight 3B-parameter text-to-speech model offering voice cloning, voice design, and voice direction via natural-language instructions, with streaming latency under 40 ms TTFA on an H100. It ships PyTorch inference code, a streaming/batch API, and benchmark suites, positioning itself as a frontier-quality open TTS alternative to proprietary systems.

Target audience: machine-learning engineers, AI researchers, and developers building voice-enabled applications

Repository: https://huggingface.co/cekal/hlas-1 · Python · Apache-2.0 · 630 stars

View on Hacker News