Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

StyleSpeech: Parameter-efficient Fine Tuning for Pre-trained Controllable Text-to-Speech

Haowei Lou, Helen Paik, Wen Hu, Lina Yao

TL;DR

StyleSpeech tackles controllable TTS by decoupling style from phoneme representations through a Style Decorator, enabling parameter-efficient style adaptation with LoRA. The architecture blends an Acoustic Pattern Encoder, Phoneme Duration Adaptor, and Griffin-Lim vocoder, while preserving phoneme identity and allowing rapid addition of new styles. A novel LLM-Guided MOS metric is proposed to objectively assess TTS quality alongside traditional metrics like $WER$, $MCD$, and PESQ. Experiments on the Baker Chinese dataset show notable improvements in both objective and perceptual measures, supporting StyleSpeech as a practical solution for dynamic, specialized TTS applications such as virtual assistants and adaptive audio content.

Abstract

This paper introduces StyleSpeech, a novel Text-to-Speech~(TTS) system that enhances the naturalness and accuracy of synthesized speech. Building upon existing TTS technologies, StyleSpeech incorporates a unique Style Decorator structure that enables deep learning models to simultaneously learn style and phoneme features, improving adaptability and efficiency through the principles of Lower Rank Adaptation~(LoRA). LoRA allows efficient adaptation of style features in pre-trained models. Additionally, we introduce a novel automatic evaluation metric, the LLM-Guided Mean Opinion Score (LLM-MOS), which employs large language models to offer an objective and robust protocol for automatically assessing TTS system performance. Extensive testing on benchmark datasets shows that our approach markedly outperforms existing state-of-the-art baseline methods in producing natural, accurate, and high-quality speech. These advancements not only pushes the boundaries of current TTS system capabilities, but also facilitate the application of TTS system in more dynamic and specialized, such as interactive virtual assistants, adaptive audiobooks, and customized voice for gaming. Speech samples can be found in https://style-speech.vercel.app

StyleSpeech: Parameter-efficient Fine Tuning for Pre-trained Controllable Text-to-Speech

TL;DR

StyleSpeech tackles controllable TTS by decoupling style from phoneme representations through a Style Decorator, enabling parameter-efficient style adaptation with LoRA. The architecture blends an Acoustic Pattern Encoder, Phoneme Duration Adaptor, and Griffin-Lim vocoder, while preserving phoneme identity and allowing rapid addition of new styles. A novel LLM-Guided MOS metric is proposed to objectively assess TTS quality alongside traditional metrics like , , and PESQ. Experiments on the Baker Chinese dataset show notable improvements in both objective and perceptual measures, supporting StyleSpeech as a practical solution for dynamic, specialized TTS applications such as virtual assistants and adaptive audio content.

Abstract

This paper introduces StyleSpeech, a novel Text-to-Speech~(TTS) system that enhances the naturalness and accuracy of synthesized speech. Building upon existing TTS technologies, StyleSpeech incorporates a unique Style Decorator structure that enables deep learning models to simultaneously learn style and phoneme features, improving adaptability and efficiency through the principles of Lower Rank Adaptation~(LoRA). LoRA allows efficient adaptation of style features in pre-trained models. Additionally, we introduce a novel automatic evaluation metric, the LLM-Guided Mean Opinion Score (LLM-MOS), which employs large language models to offer an objective and robust protocol for automatically assessing TTS system performance. Extensive testing on benchmark datasets shows that our approach markedly outperforms existing state-of-the-art baseline methods in producing natural, accurate, and high-quality speech. These advancements not only pushes the boundaries of current TTS system capabilities, but also facilitate the application of TTS system in more dynamic and specialized, such as interactive virtual assistants, adaptive audiobooks, and customized voice for gaming. Speech samples can be found in https://style-speech.vercel.app
Paper Structure (18 sections, 2 equations, 3 figures, 3 tables, 1 algorithm)

This paper contains 18 sections, 2 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. StyleSpeech
  4. Acoustic Pattern Encoder
  5. Phoneme Duration Adaptor
  6. Style Decorator
  7. Vocoder
  8. Experiments
  9. Dataset
  10. Configuration
  11. Evaluation Metrics
  12. Quantitative Metric
  13. LLM-Guided MOS
  14. Results and Discussion
  15. Imapact of Style Decorator
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Style Decorator
  • Figure 2: StyleSpeech Architecture Diagram: Figure \ref{['fig:style_speech']} presents StyleSpeech Overview. Parameters within the layers marked by blue lines are frozen during LoRA training, while parameters along the green path remain trainable, marking the area for style adaptation. "Stage N fusion" denotes the location where style and phoneme fusion occurs. Figure \ref{['fig:fusion']} shows the Style Fusion Process. Figure \ref{['fig:ape']}: shows Structures of Acoustic Feature Encoder. Figure \ref{['fig:fft']}: shows the structure of FFT Blocks.
  • Figure 3: Mel-Spectrogram diagram produced by various TTS systems when synthesizing the Pinyin phoneme Chong with different tones, Chong1, Chong2, Chong3, and Chong4. Since the values for the ground truth diagram above 30 frequency are inactive, we have cropped all values above 30 in the synthesized speech for easier comparison.