Meta-StyleSpeech : Multi-Speaker Adaptive Text-to-Speech Generation

By Dong Min, Dong Bok Lee, Eunho Yang, Sung Ju Hwang. 2021

227 citations recorded by the source corpus.

With rapid progress in neural text-to-speech (TTS) models, personalized speech generation is now in high demand for many applications. For practical applicability, a TTS model should generate high-quality speech with only a few audio samples from the given speaker, that are also short in length. However, existing methods either require to fine-tune the model or achieve low adaptation quality without fine-tuning. In this work, we propose StyleSpeech, a new TTS model which not only synthesizes high-quality speech but also effectively adapts to new speakers. Specifically, we propose Style-Adaptive Layer Normalization (SALN) which aligns gain and bias of the text input according to the style extracted from a reference speech audio. With SALN, our model effectively synthesizes speech in the style of the target speaker even from single speech audio. Furthermore, to enhance StyleSpeech's adaptation to speech from new speakers, we extend it to Meta-StyleSpeech by introducing two discriminators trained with style prototypes, and performing episodic training. The experimental results show that our models generate high-quality speech which accurately follows the speaker's voice with single short-duration (1-3 sec) speech audio, significantly outperforming baselines.

Subfields: Artificial Intelligence, Speech & Audio, Meta & Continual Learning, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xu Tan
2. Tao Qin
3. Tie-Yan Liu
4. Zhou Zhao
5. Yi Ren
6. Sheng Zhao
7. Rongjie Huang
8. Jinglin Liu

In Speech & Audio

1. Xu Tan
2. Yi Ren
3. Zhou Zhao
4. Rongjie Huang
5. Sheng Zhao
6. Jinglin Liu
7. Tao Qin
8. Tie-Yan Liu

In Meta & Continual Learning

1. Rongjie Huang
2. Zhou Zhao
3. Yi Ren
4. Jinglin Liu
5. Tao Qin
6. Zalán Borsos
7. Tie-Yan Liu
8. Sung Ju Hwang

In Signal Processing

1. Xu Tan
2. Yi Ren
3. Zhou Zhao
4. Rongjie Huang
5. Jinglin Liu
6. Tao Qin
7. Tie-Yan Liu
8. Sheng Zhao

Citations in the corpus

• Grad-StyleSpeech: Any-Speaker Adaptive Text-to-Speech Synthesis with Diffusion Models (2022, methodology): “In contrast to previous diffusion-based approaches [7, 8, 9], we construct the model using a style-based generative model [15] to take the target speaker’s style into account.”
• A Survey on Neural Speech Synthesis (2021, background): “Efficient Adaptation Few-Data Adaptation [43, 9, 173, 234, 439, 48, 39, 230] Untranscribed Data Adaptation [397, 130, 215] Few-Parameter Adaptation [9, 43, 39] Zero-Shot Adaptation [9, 43, 139, 55]”
• InstructTTS: Modelling Expressive TTS in Discrete Latent Space With Natural Language Style Prompt (2023, background): “In the field of expressive speech synthesis, several works [6], [9], [12], [27], [28], [29] have been done.”
• ProDiff: Progressive Fast Diffusion Model for High-Quality Text-to-Speech (2022, methodology): “Following the common practice [1, 28], we conduct preprocessing on the speech and text data: 1) extract the spectrogram with the FFT size of 1024, hop size of 256, and window size of 1024 samples; 2) convert it to a mel-spectrogram with 80 frequency bins; and 3) extract F0 (fundamental frequency) fr”
• Make-A-Voice: Unified Voice Synthesis With Discrete Representation (2023, methodology): “We compare the generated audio samples with other systems, including 1) GT, the ground-truth audio; 2) Meta-StyleSpeech [27]: the multi-speaker model fine-tuned with meta-learning; 3) GenerSpeech, a generalizable model with the global and local stylization for speaker and prosody style transfer; 4)…”
• Multimodal Learning With Transformers: A Survey (2022, methodology): “[177] D. Min, D. B. Lee, E. Yang, and S. J. Hwang, “Meta-StyleSpeech: Multispeaker adaptive text-to-speech generation,” 2021, arXiv:2106.03153.”
• Mega-TTS 2: Zero-Shot Text-to-Speech with Arbitrary Length Speech Prompts (2023, methodology): “, Meta-stylespeech (Min et al., 2021), Gradstylespeech (Kang et al.”
• DiffProsody: Diffusion-Based Latent Prosody Generation for Expressive Speech Synthesis With Prosody Conditional Adversarial Training (2023, methodology): “Meta-StyleSpeech [11] proposes the application of style”
• Nonparallel Expressive TTS for Unseen Target Speaker using Style-Controlled Adaptive Layer and Optimized Pitch Embedding (2023, methodology): “Min et al. [10] developed Meta-StyleSpeech, which employs a meta-learning approach for TTS synthesis with multiple speakers.”
• RetrieverTTS: Modeling Decomposed Factors for Text-Based Speech Insertion (2022, methodology): “We conduct a similarity test to further confirm the superi-Table 2, we observe a significant SMOS drop when replacing our global factor extraction and injection method with that proposed in [12].”
• StyleTTS: A Style-Based Generative Model for Natural and Diverse Text-to-Speech Synthesis (2022, unknown): “Application of these methods in speech synthesis has not been extensivelyexploredyet,restrictedprimarilytovoiceconversion andspeakeradaptation[32],[33],[34],[35],[36].”
• SingGAN: Generative Adversarial Network For High-Fidelity Singing Voice Generation (2021, background): “, 16k or 22kHz) in most scenarios [3, 18], singing voice recordings usually have a higher sampling rate (e.”
• GenerSpeech: Towards Style Transfer for Generalizable Out-Of-Domain Text-to-Speech Synthesis (2022, methodology): “We conduct crowd-sourced human evaluations with MOS (mean opinion score) for naturalness and SMOS (similarity mean opinion score) [31] for style similarity on the testing set.”
• SpeechPainter: Text-conditioned Speech Inpainting (2022, methodology): “We rely on Meta-StyleSpeech [14] for adaptive TTS, using the pretrained model provided by the authors2 and a vocoder matching Meta-StyleSpeech’s frontend trained in the same manner as SpeechPainter’s vocoder (Section 2.”
• M2-CTTS: End-to-End Multi-Scale Multi-Modal Conversational Text-to-Speech Synthesis (2023, methodology): “emotional speech corpus, many expressive speech synthesis systems have been proposed and they [3, 4, 5, 6, 7] can produce speech with rich prosody and expressiveness.”
• HierSpeech++: Bridging the Gap Between Semantic and Acoustic Representation of Speech by Hierarchical Variational Inference for Zero-Shot Speech Synthesis (2023, unknown): “Multi-SpectroGAN [53] used adversarial feedback and a mixup strategy for an expressive and diverse zero-shot TTS. Meta-StyleSpeech [54] introduced meta-learning for style modeling, and GenerSpeech [55] used a mix-style layer normalization for better generalization on the out-of-domain style…”