analiza aiqbench
Tacit-TTS addresses the latency bottleneck in autoregressive text-to-speech systems by switching to masked non-autoregressive generation, achieving over 10x speedup for long utterances without requiring reference transcripts. This advancement enables efficient zero-shot voice cloning that remains robust against unreliable ASR outputs and supports cross-lingual synthesis using non-lexical references like infant babble. The work represents a significant step forward in balancing high-fidelity voice cloning with practical inference speeds for real-world applications.
Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning
arXiv:2609.38658v1 Announce Type: cross Abstract: 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. We present Tacit-TTS, an efficient transcript-free zero-shot voice cloning system distilled from IndexTTS2. Our model replaces autoregressive text-to-semantic decoding with masked non-autoregressive generation, introduces training-free acoustic length estimation, and accelerates the flow-matching renderer through ReFlow distillation. Across two English and two Mandarin datasets, Tacit-TTS achieves competitive zero-shot quality while generating speech over 10x faster than IndexTTS2 for utterances longer than 5 seconds. Its transcript-free conditioning further supports cross-lingual and non-lexical references. We validate this capability using references from eight other languages, infant babble, and synthetic gibberish, where transcript-dependent systems often degrade or fail due to unreliable ASR transcripts.
Abstract: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. We present Tacit-TTS, an efficient transcript-free zero-shot voice cloning system distilled from IndexTTS2. Our model replaces autoregressive text-to-semantic decoding with masked non-autoregressive generation, introduces training-free acoustic length estimation, and accelerates the flow-matching renderer through ReFlow distillation. Across two English and two Mandarin datasets, Tacit-TTS achieves competitive zero-shot quality while generating speech over 10x faster than IndexTTS2 for utterances longer than 5 seconds. Its transcript-free conditioning further supports cross-lingual and non-lexical references. We validate this capability using references from eight other languages, infant babble, and synthetic gibberish, where transcript-dependent systems often degrade or fail due to unreliable ASR transcripts.
Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Sound (cs.SD)
arXiv:2609.38658 [eess.AS]
(or arXiv:2609.38658v1 [eess.AS] for this version)
https://doi.org/10.48550/arXiv.2609.38658 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 248 cuvinte
Subiecte conexe
Parallelize speculative decoding with P-EAGLE on Amazon SageMaker AI
AWS · 16 iun. 2026
HALO: Enhancing Time Series Generation via Hyperspherical Latents and Masked AutoregRessive Modeling
arXiv · 29 sept. 2026
LESS Is More: Mutual-Stability Sampling for Diffusion Language Models
arXiv · 17 iun. 2026
Flow Reasoning Models: Turning Discrete Flows Into Efficient Recurrent Reasoners
arXiv · 1 sept. 2026