Speaker-Dependent WaveNet Vocoder

By Akira Tamamori, Tomoki Hayashi, Kazuhiro Kobayashi, Kazuya Takeda, Tomoki Toda. 2017

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

In Speech & Audio

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

In Signal Processing

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

Citations in the corpus

• A Survey on Neural Speech Synthesis (2021, methodology): “The original WaveNet, as well as some following works that leverage WaveNet as vocoder [8, 84], generate speech waveform conditioned on linguistic features, while WaveNet can be easily adapted to condition on linear-spectrograms [84] and mel-spectrograms [330, 264, 297].”
• TTS-Portuguese Corpus: a corpus for speech synthesis in Brazilian Portuguese (2020, methodology): “, 2018) combines Tacotron 1 with a modified WaveNet vocoder (Tamamori et al., 2017).”
• Wavenet Based Low Rate Speech Coding (2017, background): “The parametric coder model differs from [14] in that it is not speaker dependent, and that it can also be decoded with a conventional low complexity decoder.”
• Probability density distillation with generative adversarial networks for high-quality parallel waveform generation (2019, methodology): “ter size was 3. The conditioning features were upsampled by nearest neighbor upsampling followed by 2-D convolutions for interpolation [32]. The upsampling was split into five modules. The scales were [2, 2, 2, 3, 5]. The kernel sizes for the 2-D convolutions were set to 2 s+ 1, where denotes the ups”
• Neural Source-Filter Waveform Models for Statistical Parametric Speech Synthesis (2019, methodology): “The pioneering model, called “WaveNet-vocoder” [12], adopts a dilated convolution network [13] to directly generate waveform samples given the input acoustic features.”
• Neural Source-filter-based Waveform Model for Statistical Parametric Speech Synthesis (2018, methodology): “Compared with the AR WaveNet, the proposed model generated speech with a similar quality at a much faster speed.”
• Improving Prosody with Linguistic and Bert Derived Features in Multi-Speaker Based Mandarin Chinese Neural TTS (2020, methodology): “Through refining model structure and cascading with a modified WaveNet vocoder [12][13], audios generated by Tacotron2 [14] has shown high TTS quality, which is close to natural voice of human.”
• A Vector Quantized Variational Autoencoder (VQ-VAE) Autoregressive Neural $F_0$ Model for Statistical Parametric Speech Synthesis (2020, methodology): “Given the predicted F0 and spectral features, SPSS systems generate a speech waveform using a deterministic or neural vocoder [10], [11].”
• Semi-Supervised Speaker Adaptation for End-to-End Speech Synthesis with Pretrained Models (2020, methodology): “In waveform synthesis, WaveNet [30] conditioned by 80-band log Mel-filter bank features was used as a vocoder [31].”
• An Investigation of Subband Wavenet Vocoder Covering Entire Audible Frequency Range with Limited Acoustic Features (2018, methodology): “In addition, to fuel conventional source-filter model-based vocoders within a raw audio generative model framework, a WaveNet vocoder has been proposed [18], which directly synthesizes raw speech waveforms from acoustic features, and applied to a conventional VC framework [19].”
• Wasserstein GAN and Waveform Loss-Based Acoustic Model Training for Multi-Speaker Text-to-Speech Synthesis Systems Using a WaveNet Vocoder (2018, methodology): “In other studies [14], [15], [27], only acoustic features are used as the local con-”
• Controllable Neural Prosody Synthesis (2020, methodology): “We use a publicly available implementation [30] of a single-speaker WaveNet vocoder [31] and predict 10-bit μ-law-encoded waveform samples.”
• End-To-End Speech Synthesis Applied to Brazilian Portuguese (2020, methodology): “Tacotron 2 [Shen et al., 2018] combines Tacotron 1 with modified wavenet vocoder [Tamamori et al., 2017b].”
• Speaker-independent raw waveform model for glottal excitation (2018, methodology): “In the statistical vocoder approach, a WaveNet is conditioned with some acoustic features, such as mel filterbank energies [2, 3], or mel-generalized cepstrum (MGC) coefficients and F0 [4].”
• Improving FFTNet Vocoder with Noise Shaping and Subband Approaches (2018, methodology): “Thirty layers (10 dilations × 3 cycles) with a kernel size of 2 were used for the dilated causal convolution layers, giving a receptive field of 3070 samples [17,28].”
• Fftnet: A Real-Time Speaker-Dependent Neural Vocoder (2018, methodology): “This result also contradicts MCD and RMSE.”