Thousands of Voices for HMM-Based Speech Synthesis–Analysis and Application of TTS Systems Built on Various ASR Corpora

By Junichi Yamagishi, Bela Usabaev, Simon King, Oliver Watts, John Dines, Jilei Tian, Yong Guan, Rile Hu, Keiichiro Oura, Yi-Jian Wu and 3 more. 2010

102 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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. Junichi Yamagishi
5. Sheng Zhao
6. Frank K. Soong
7. Heiga Zen
8. Yi Ren

In Signal Processing

1. Junichi Yamagishi
2. Xu Tan
3. Tao Qin
4. Tie-Yan Liu
5. Sheng Zhao
6. Heiga Zen
7. Tomoki Toda
8. Yi Ren

Citations in the corpus

• Deep Voice 3: 2000-Speaker Neural Text-to-Speech (2017, background): “For example, Yamagishi et al. (2010) builds several speaker-adaptive HMM-based TTS systems (Yam-agishi et al., 2009) on various ASR corpora with hundreds of speakers.”
• Deep Voice 3: Scaling Text-to-Speech with Convolutional Sequence Learning (2017, methodology): “Although prior work has applied TTS methods to ASR datasets (Yamagishi et al., 2010), Deep Voice 3 is, to the best of our knowledge, the first TTS system to scale to thousands of speakers with a single model.”
• Training Multi-Speaker Neural Text-to-Speech Systems using Speaker-Imbalanced Speech Corpora (2019, methodology): “In another approach, we could combine speech data from many speakers and train a multi-speaker TTS system [9].”
• Speech Synthesis Based on Hidden Markov Models (2013, background): “Speaker-adaptive HMMbased synthesis requires as little as 5–7 min of recorded speech from a target speaker to generate a personalized synthetic voice [24]; hence, the average voice model can be easily transformed into a synthetic voice for any number of speakers [25].”
• The voice bank corpus: Design, collection and data analysis of a large regional accent speech database (2013, methodology): “This technique, referred to as speaker adaptation, has already been successfully applied to “non traditional TTS” corpora such as ASR corpora, which contains a large number of speakers [4].”
• Utterance Selection Techniques for TTS Systems Using Found Speech (2016, methodology): “Previous methods have looked at building speech synthesis systems from audio books [3],[4] and ASR corpora [5].”
• Pretraining Strategies, Waveform Model Choice, and Acoustic Configurations for Multi-Speaker End-to-End Speech Synthesis (2020, background): “Conversely, [9] trained a base model on noisy speech and adapted to clean recordings, finding that the inclusion of noisy data resulted in improvements over just using the clean data by itself, when only a small amount of clean data is available.”
• LRSpeech: Extremely Low-Resource Speech Synthesis and Recognition (2020, background): “arget-speaker voice and improve the ASR model on multiple voices. 1.3 Data Cost and Accuracy Next, we introduce the extremely low data cost while promising accuracy achieved by LRSpeech. According to [4, 14, 15, 38, 43], the pronunciation lexicon, singlespeaker high-quality paired data and single-sp”