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

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...

Full description

Saved in:
Bibliographic Details
Main Authors: Yeunju Choi, Youngmoon Jung, Youngjoo Suh, Hoirin Kim
Format: Article
Language:English
Published: IEEE 2022-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/9775804/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850190900438761472
author Yeunju Choi
Youngmoon Jung
Youngjoo Suh
Hoirin Kim
author_facet Yeunju Choi
Youngmoon Jung
Youngjoo Suh
Hoirin Kim
author_sort Yeunju Choi
collection DOAJ
description 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.
format Article
id doaj-art-94c605add4db4b9fa8fb5f63361d4929
institution OA Journals
issn 2169-3536
language English
publishDate 2022-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-94c605add4db4b9fa8fb5f63361d49292025-08-20T02:15:07ZengIEEEIEEE Access2169-35362022-01-0110526215262910.1109/ACCESS.2022.31758109775804Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-SpeechYeunju Choi0https://orcid.org/0000-0003-2192-2680Youngmoon Jung1https://orcid.org/0000-0002-4321-379XYoungjoo Suh2Hoirin Kim3https://orcid.org/0000-0002-8787-6982School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South KoreaSchool of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South KoreaVoice Group, Konan Technology Inc., Seoul, South KoreaSchool of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South KoreaAlthough 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.https://ieeexplore.ieee.org/document/9775804/MOS predictionneural TTSperceptual lossspeech synthesis
spellingShingle Yeunju Choi
Youngmoon Jung
Youngjoo Suh
Hoirin Kim
Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
IEEE Access
MOS prediction
neural TTS
perceptual loss
speech synthesis
title Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
title_full Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
title_fullStr Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
title_full_unstemmed Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
title_short Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
title_sort learning to maximize speech quality directly using mos prediction for neural text to speech
topic MOS prediction
neural TTS
perceptual loss
speech synthesis
url https://ieeexplore.ieee.org/document/9775804/
work_keys_str_mv AT yeunjuchoi learningtomaximizespeechqualitydirectlyusingmospredictionforneuraltexttospeech
AT youngmoonjung learningtomaximizespeechqualitydirectlyusingmospredictionforneuraltexttospeech
AT youngjoosuh learningtomaximizespeechqualitydirectlyusingmospredictionforneuraltexttospeech
AT hoirinkim learningtomaximizespeechqualitydirectlyusingmospredictionforneuraltexttospeech