Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning
eess.AS, cs.CL, cs.SD
Submitted: 2026-09-29
Updated: 2026-09-29
Comments: Under Review
Code: https://github.com/textstat/textstat
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
The gist: 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.
Terminology
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.
Sources
- Seed-TTS: A Family of High-Quality Versatile Speech Generation Models
- Seamless: Multilingual Expressive and Streaming Speech Translation
- SoundStorm: Efficient Parallel Audio Generation
- IndexTTS: An Industrial-Level Controllable and Efficient Zero-Shot Text-To-Speech System
- CosyVoice: A Scalable Multilingual Zero-shot Text-to-speech Synthesizer based on Supervised Semantic Tokens
- CosyVoice 2: Scalable Streaming Speech Synthesis with Large Language Models
- FunASR: A Fundamental End-to-End Speech Recognition Toolkit
- Distilling the Knowledge in a Neural Network
- BigVGAN: A Universal Neural Vocoder with Large-Scale Training
- Flow Matching for Generative Modeling
- Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow
- FastSpeech 2: Fast and High-Quality End-to-End Text to Speech
- UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
- Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers
- CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking
- Spark-TTS: An Efficient LLM-Based Text-to-Speech Model with Single-Stream Decoupled Speech Tokens
Related papers
- Rethinking Entropy Allocation in LLM-based ASR: Understanding the Dynamics between Speech Encoders and LLMs
- X-VC: Zero-shot Streaming Voice Conversion in Codec Space
- Autoregressive Guidance of Deep Spatially Selective Filters using Bayesian Tracking for Efficient Extraction of Moving Speakers
- VM-ARRAYDPS: Virtual Microphone Augmented Diffusion Posterior Sampling for Unsupervised Blind Speech Separation
- Conditional Flow Matching for Visually-Guided Acoustic Highlighting
- Anonymization, Not Elimination: Utility-Preserved Speech Anonymization