Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning
Tacit-TTS distills IndexTTS2 into transcript-free masked voice cloning that is over 10x faster.
Tacit-TTS is a transcript-free zero-shot voice-cloning system distilled from IndexTTS2. It replaces autoregressive text-to-semantic decoding with masked non-autoregressive generation, estimates acoustic length without extra training, and accelerates the flow-matching renderer with ReFlow distillation. On two English and two Mandarin datasets it stays competitive while generating speech over 10x faster than IndexTTS2 for utterances longer than 5 seconds. References from eight other languages, infant babble, and synthetic gibberish still work where transcript-dependent systems fail.
- Distilled from IndexTTS2 using masked non-autoregressive generation
- Adds training-free length estimation and ReFlow flow-matching distillation
- Over 10x faster than IndexTTS2 for utterances longer than 5 seconds
- Transcript-free cloning works across languages, babble, and gibberish
From the original article
1 min read at huggingface.co
TTS systems with autoregressive semantic modeling have demonstrated strong zero-shot voice cloning performance and rich expressive variation, but their sequential decoding incurs substantial latency. Non-autoregressive alternatives offer much faster generation, yet often rely on more restrictive reference conditioning, such as requiring transcripts of the reference speech during inference.
By Jian Chen, You Zhang, Mark Vinton · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.