Speech Synthesis Based on Hidden Markov Models

By Keiichi Tokuda, Yoshihiko Nankaku, Tomoki Toda, Heiga Zen, Junichi Yamagishi, Keiichiro Oura. 2013

468 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio, 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. Frank K. Soong
6. Sheng Zhao
7. Lei He
8. Yu Zhang

In Speech & Audio

1. Xu Tan
2. Junichi Yamagishi
3. Tao Qin
4. Tie-Yan Liu
5. Sheng Zhao
6. Frank K. Soong
7. Lei He
8. Tomoki Toda

In Signal Processing

1. Junichi Yamagishi
2. Xu Tan
3. Tao Qin
4. Frank K. Soong
5. Tie-Yan Liu
6. Tomoki Toda
7. Shinji Watanabe
8. Sheng Zhao

Citations in the corpus

• Neural Speech Synthesis with Transformer Network (2018, methodology): “Parametric TTS systems (Tokuda et al. 2000; Zen, Tokuda, and Black 2009; Ze, Senior, and Schuster 2013; Tokuda et al. 2013) convert speech waves into spectrograms, and acoustic parameters, such as fundamental frequency and duration, are used to synthesize new audio results.”
• A Survey on Neural Speech Synthesis (2021, methodology): “One major drawback of HMM-based SPSS is that the quality of the synthesized speech is not good enough [418, 351], mainly due to two reasons: 1) the accuracy of acoustic models is not good, and the predicted acoustic features are oversmoothing and lack of details, and 2) the vocoding techniques are n”
• SampleRNN: An Unconditional End-to-End Neural Audio Generation Model (2016, neutral): “Chung et al. (2015) also attempt to model raw audio waveforms which is in contrast to traditional approaches which use spectral features as in Tokuda et al. (2013), Bertrand et al. (2008), and Lee et al. (2009).”
• Glow-TTS: A Generative Flow for Text-to-Speech via Monotonic Alignment Search (2020, methodology): “Traditionally, hidden Markov models (HMMs) have been used to estimate unknown alignments between text and speech [19, 25].”
• Adversarial Audio Synthesis (2018, neutral): “Some of these systems use learning-based approaches such as a hidden Markov models (Yoshimura, 2002; Tokuda et al., 2013), and separately-trained neural networks pipelines (Ling et al., 2015) to estimate speech parameters.”
• Neural Source-Filter Waveform Models for Statistical Parametric Speech Synthesis (2019, methodology): “In TTS systems based on a statistical parametric speech synthesis (SPSS) framework [2], [3], deep neural networks have been used to derive linguistic features from text [4].”
• Discovering linguistic structures in speech: models and applications (2014, methodology): “While it is easy to forward sample MFCC features from the whole-word HMM, we adopt the procedure used by most HMM-based speech synthesis systems [178, 189], and generate the mean vector of the most weighted Gaussian mixture component for each sub-state in the whole-word HMM.”
• Semi-Supervised Speaker Adaptation for End-to-End Speech Synthesis with Pretrained Models (2020, background): “Differing from conventional statistical parametric speech synthesis (SPSS) systems based on hidden Markov models (HMMs) [6] and deep neural networks (DNNs) [7], end-to-end TTS systems can be trained with raw text data directly without requiring linguistic features, e.g., phoneme, syllable, accent,…”
• Aligntts: Efficient Feed-Forward Text-to-Speech System Without Explicit Alignment (2020, methodology): “Reviewing the traditional statistical parametric speech synthesis system based on the Hidden Markov Models (HMMs) [15, 16], the Baum-Welch algorithm [17] is used to estimate the parameters of HMMs if the alignment between states and observations is unknown.”
• Statistical Parametric Speech Synthesis Incorporating Generative Adversarial Networks (2017, methodology): “features and acoustic features more accurately than conventional hidden Markov models [5] and Gaussian mixture models [6].”
• A Neural Vocoder With Hierarchical Generation of Amplitude and Phase Spectra for Statistical Parametric Speech Synthesis (2019, background): “Hidden Markov models (HMMs) [1], deep neural networks (DNNs) [2],”
• Generative Adversarial Network-Based Glottal Waveform Model for Statistical Parametric Speech Synthesis (2017, background): “SPSS systems have several advantages over concatenative synthesis, such as their flexibility to transform the synthesis to different voice characteristics, speaking styles and emotions, as well as their small memory footprint and robustness to unseen text prompts [1].”
• Merlin: An Open Source Neural Network Speech Synthesis System (2016, background): “For the past decade, hidden Markov models (HMMs) have dominated acoustic modelling [4].”
• An investigation of multi-speaker training for wavenet vocoder (2017, methodology): “Moreover, it is straightforward to apply a statistical framework to this approach [13].”
• Expressive TTS Training With Frame and Style Reconstruction Loss (2020, background): “Studies on hierarchical modeling of F0 in speech synthesis [1],”
• A Review of Deep Learning Based Speech Synthesis (2019, methodology): “…of speech synthesis technologies, from the previous formant based parametric synthesis [1,2], waveform concatenation based methods [3–5] to the current statistical parametric speech synthesis (SPSS) [6], the intelligibility and naturalness of the synthesized speech have been improved greatly.”