AI
Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 10:21:47 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 19:18:24 UTC.
Briefings
Tacit-TTS is a transcript-free voice cloning system distilled from IndexTTS2 that replaces autoregressive decoding with masked non-autoregressive generation. It achieves competitive zero-shot quality across two English and two Mandarin datasets while generating speech over 10x faster than IndexTTS2 for utterances longer than 5 seconds. The system supports cross-lingual and non-lexical references by using references from eight other languages, infant babble, and synthetic gibberish, where transcript-dependent systems often fail. The model introduces training-free acoustic length estimation and accelerates the flow-matching renderer through ReFlow distillation.
Key points
- Tacit-TTS replaces autoregressive text-to-semantic decoding with masked non-autoregressive generation to achieve transcript-free voice cloning.
- The system generates speech over 10x faster than IndexTTS2 for utterances longer than 5 seconds across two English and two Mandarin datasets.
Context and limitations
Performance and quality are validated only on two English and two Mandarin datasets; transcript-free conditioning may degrade or fail on references from other languages or when ASR transcripts are unreliable.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 10:21:47 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihmgrx362jmggd47gtjxtjjdgrc5ajz2efxet5uoijmaqpapghyuq