Improved parallel WaveGAN vocoder with perceptually weighted spectrogram loss

01/19/2021

∙

This paper proposes a spectral-domain perceptual weighting technique for Parallel WaveGAN-based text-to-speech (TTS) systems. The recently proposed Parallel WaveGAN vocoder successfully generates waveform sequences using a fast non-autoregressive WaveNet model. By employing multi-resolution short-time Fourier transform (MR-STFT) criteria with a generative adversarial network, the light-weight convolutional networks can be effectively trained without any distillation process. To further improve the vocoding performance, we propose the application of frequency-dependent weighting to the MR-STFT loss function. The proposed method penalizes perceptually-sensitive errors in the frequency domain; thus, the model is optimized toward reducing auditory noise in the synthesized speech. Subjective listening test results demonstrate that our proposed method achieves 4.21 and 4.26 TTS mean opinion scores for female and male Korean speakers, respectively.

READ FULL TEXT

Improved parallel WaveGAN vocoder with perceptually weighted spectrogram loss

Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram

Parallel waveform synthesis based on generative adversarial networks with voicing-aware conditional discriminators

Probability density distillation with generative adversarial networks for high-quality parallel waveform generation

Speech waveform synthesis from MFCC sequences with generative adversarial networks

Attention-Guided Generative Adversarial Network for Whisper to Normal Speech Conversion

Quasi-Periodic Parallel WaveGAN: A Non-autoregressive Raw Waveform Generative Model with Pitch-dependent Dilated Convolution Neural Network

Improving the expressiveness of neural vocoding with non-affine Normalizing Flows

Improved parallel WaveGAN vocoder with perceptually weighted spectrogram loss

Related Research

Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram

Parallel waveform synthesis based on generative adversarial networks with voicing-aware conditional discriminators

Probability density distillation with generative adversarial networks for high-quality parallel waveform generation

Speech waveform synthesis from MFCC sequences with generative adversarial networks

Attention-Guided Generative Adversarial Network for Whisper to Normal Speech Conversion

Quasi-Periodic Parallel WaveGAN: A Non-autoregressive Raw Waveform Generative Model with Pitch-dependent Dilated Convolution Neural Network

Improving the expressiveness of neural vocoding with non-affine Normalizing Flows