Recent Advances in Google Real-Time HMM-Driven Unit Selection Synthesizer

By X. Gonzalvo, Siamak Tazari, Chun-an Chan, Markus Becker, Alexander Gutkin, Hanna Silén. 2016

61 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Hardware and Architecture, Natural Language 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. Yu Zhang
2. Yonghui Wu
3. Jinyu Li
4. Tao Qin
5. Xu Tan
6. Tie-Yan Liu
7. Dong Yu
8. Junichi Yamagishi

In Hardware and Architecture

1. Bo Li
2. Quoc V. Le
3. Patrick Nguyen
4. Samy Bengio
5. Alex Graves
6. Zongheng Yang
7. Z. Chen
8. Navdeep Jaitly

In Natural Language Processing

1. Yu Zhang
2. Yoshua Bengio
3. Yonghui Wu
4. Dong Yu
5. Quoc V. Le
6. Tara N. Sainath
7. Tao Qin
8. Tie-Yan Liu

Citations in the corpus

• WaveNet: A Generative Model for Raw Audio (2016, methodology): “As example-based and model-based speech synthesis baselines, hidden Markov model (HMM)-driven unit selection concatenative (Gonzalvo et al., 2016) and long short-term memory recurrent neural network (LSTM-RNN)-based statistical parametric (Zen et al.”
• Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions (2017, neutral): “We also compare to the original Tacotron that predicts linear spectrograms and uses Griffin-Lim to synthesize audio, as well as concatenative [30] and parametric [31] baseline systems, both of which have been used in production at Google.”
• Tacotron: Towards End-to-End Speech Synthesis (2017, methodology): “We compare our model with a parametric (based on LSTM (Zen et al., 2016)) and a concatenative system (Gonzalvo et al., 2016), both of which are in production.”
• Deep Voice 3: 2000-Speaker Neural Text-to-Speech (2017, background): “2 JOINT REPRESENTATION OF CHARACTERS AND PHONEMES Deployed TTS systems (e.g., Capes et al., 2017; Gonzalvo et al., 2016) should include a way to modify pronunciations to correct common mistakes (which typically include proper nouns, foreign words, and domain-specific jargon).”
• Fast, Compact, and High Quality LSTM-RNN Based Statistical Parametric Speech Synthesizers for Mobile Devices (2016, methodology): “Statistical parametric speech synthesis (SPSS) [1] based on artificial neural networks (ANN) has became popular in the text-to-speech (TTS) research area in the last few years [2–20].”
• Streaming End-to-end Speech Recognition for Mobile Devices (2018, methodology): “We synthesize this data using a concatenative TTS approach with one voice [43] to create audio-text pairs, which we augment to our training data (each batch is filled with supervised data 90% of the time and synthetic data 10% of the time).”
• Siri On-Device Deep Learning-Guided Unit Selection Text-to-Speech System (2017, methodology): “The classical unit selection approaches typically use either rulebased methods, hidden Markov models (HMMs) or neural networks (NN) for target cost [2, 8, 4, 5, 6, 7, 9], and acoustic distances for concatenation cost.”
• Deep Voice 3: Scaling Text-to-Speech with Convolutional Sequence Learning (2017, background): “2 JOINT REPRESENTATION OF CHARACTERS AND PHONEMES Deployed TTS systems (e.g., Capes et al., 2017; Gonzalvo et al., 2016) should include a way to modify pronunciations to correct common mistakes (which typically involve proper nouns, foreign words, and domain-specific jargon).”
• Recognizing Long-Form Speech Using Streaming End-to-End Models (2019, methodology): “To further test generalization to long utterance lengths, we create a TTS-Audiobook test set by synthesizing passages containing at least 200 words from a novel [41] using a parametric text-to-speech (TTS) system [42], with a single voice.”
• Speaker-Dependent WaveNet Vocoder (2017, background): “In the original literature [13], it was applied to text-to-speech (TTS), and the quality of synthesized speech exceeded that of state-of-the-art approaches [14, 15].”
• Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model (2017, methodology): “We compare our model with a parametric (based on LSTM (Zen et al., 2016)) and a concatenative system (Gonzalvo et al., 2016), both of which are in production.”
• RNN-T Models Fail to Generalize to Out-of-Domain Audio: Causes and Solutions (2020, methodology): “Our out-of-domain test set consists of 7 hours of speech generated using a text-to-speech system [27], which is acoustically simple but much longer than the training utterances (TTSAudiobook; median length is 62 seconds).”
• Review of end-to-end speech synthesis technology based on deep learning (2021, methodology): “of TTS techniques from the formant-based methods [92, 97, 105, 106, 179, 224] to the unit selection-based waveform cascade methods [6, 24, 36, 59, 80, 143, 144], and to the hidden Markov model (HMM)-based statistical parametric speech synthesis (SPSS) methods [26, 94, 149, 178, 210, 241, 249, 250].”
• Two-Pass End-to-End Speech Recognition (2019, unknown): “one voice [25].”