Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech

Abstract

Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve speech quality by training a TTS model under the supervision of perceptual loss, which measures the distance between the maximum possible speech quality score and the predicted one. We first pre-train a mean opinion score (MOS) prediction model and then train a TTS model to maximize the MOS of synthesized speech using the pre-trained MOS prediction model. The proposed method can be applied independently regardless of the TTS model architecture or the cause of speech quality degradation and efficiently without increasing the inference time or model complexity. The evaluation results for the MOS and phone error rate demonstrate that our proposed approach improves previous models in terms of both naturalness and intelligibility.

9 pages, 5 figures, 4 tables

References in corpus (11)

• WaveNet: A Generative Model for Raw Audio
• MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis
• FastSpeech: Fast, Robust and Controllable Text to Speech
• Parallel WaveNet: Fast High-Fidelity Speech Synthesis
• Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention
• MOSNet: Deep Learning based Objective Assessment for Voice Conversion
• On Loss Functions for Supervised Monaural Time-Domain Speech Enhancement
• Learning with Learned Loss Function: Speech Enhancement with Quality-Net to Improve Perceptual Evaluation of Speech Quality
• Deep MOS Predictor for Synthetic Speech Using Cluster-Based Modeling
• Deep Long Audio Inpainting
• An ASR Guided Speech Intelligibility Measure for TTS Model Selection