Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Decoding for Discrete Speech Synthesis with Multi-Resolution Spoof Detection

Junchuan Zhao, Minh Duc Vu, Ye Wang

TL;DR

This work proposes MSpoof-TTS, a training-free inference framework that improves zero-shot synthesis through multi-resolution spoof guidance and introduces a Multi-Resolution Token-based Spoof Detection framework that evaluates codec sequences at different temporal granularities to detect locally inconsistent or unnatural patterns.

Abstract

Neural codec language models enable high-quality discrete speech synthesis, yet their inference remains vulnerable to token-level artifacts and distributional drift that degrade perceptual realism. Rather than relying on preference optimization or retraining, we propose MSpoof-TTS, a training-free inference framework that improves zero-shot synthesis through multi-resolution spoof guidance. We introduce a Multi-Resolution Token-based Spoof Detection framework that evaluates codec sequences at different temporal granularities to detect locally inconsistent or unnatural patterns. We then integrate the spoof detectors into a hierarchical decoding strategy, progressively pruning low-quality candidates and re-ranking hypotheses. This discriminator-guided generation enhances robustness without modifying model parameters. Experiments validate the effectiveness of our framework for robust and high-quality codec-based speech generation.

Hierarchical Decoding for Discrete Speech Synthesis with Multi-Resolution Spoof Detection

TL;DR

This work proposes MSpoof-TTS, a training-free inference framework that improves zero-shot synthesis through multi-resolution spoof guidance and introduces a Multi-Resolution Token-based Spoof Detection framework that evaluates codec sequences at different temporal granularities to detect locally inconsistent or unnatural patterns.

Abstract

Neural codec language models enable high-quality discrete speech synthesis, yet their inference remains vulnerable to token-level artifacts and distributional drift that degrade perceptual realism. Rather than relying on preference optimization or retraining, we propose MSpoof-TTS, a training-free inference framework that improves zero-shot synthesis through multi-resolution spoof guidance. We introduce a Multi-Resolution Token-based Spoof Detection framework that evaluates codec sequences at different temporal granularities to detect locally inconsistent or unnatural patterns. We then integrate the spoof detectors into a hierarchical decoding strategy, progressively pruning low-quality candidates and re-ranking hypotheses. This discriminator-guided generation enhances robustness without modifying model parameters. Experiments validate the effectiveness of our framework for robust and high-quality codec-based speech generation.
Paper Structure (15 sections, 3 figures, 3 tables, 2 algorithms)

This paper contains 15 sections, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Method
  3. Overview
  4. Multi-Resolution Token-Level Spoof Detection
  5. Hierarchical Spoof-Guided Sampling
  6. Experiments
  7. Datasets
  8. Implementation Details
  9. Evaluation Metrics
  10. Results
  11. Evaluation on Standard Benchmarks
  12. Evaluation under Challenging Conditions
  13. Subjective Listening Tests
  14. Analysis of the Multi-Resolution Spoof Detectors
  15. Conclusion

Figures (3)

  • Figure 1: Overview of multi-resolution token-based spoof detection framework. (a) Construction of token sequences at multiple temporal resolutions for training separate real/fake detectors. (b) Conformer-based discrete token spoof detector architecture.
  • Figure 2: t-SNE visualization of embedding distributions under different segment lengths.
  • Figure 3: Subjective evaluation of different inference strategies measured by MOS-N (naturalness), MOS-Q (quality), and SMOS (similarity).