• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.39-48
• /
• 2018

A typical statistical parametric speech synthesis (text-to-speech, TTS) system consists of separate modules, such as a text analysis module, an acoustic modeling module, and a speech synthesis module. This causes two problems: 1) expert knowledge of each module is required, and 2) errors generated in each module accumulate passing through each module. An end-to-end TTS system could avoid such problems by synthesizing voice signals directly from an input string. In this study, we implemented an end-to-end Korean TTS system using Google's Tacotron, which is an end-to-end TTS system based on a sequence-to-sequence model with attention mechanism. We used 4392 utterances spoken by a Korean female speaker, an amount that corresponds to 37% of the dataset Google used for training Tacotron. Our system obtained mean opinion score (MOS) 2.98 and degradation mean opinion score (DMOS) 3.25. We will discuss the factors which affected training of the system. Experiments demonstrate that the post-processing network needs to be designed considering output language and input characters and that according to the amount of training data, the maximum value of n for n-grams modeled by the encoder should be small enough.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.5
• /
• pp.614-620
• /
• 2019

In this paper, we propose a method to effectively determine the representative style embedding of each emotion class to improve the global style token-based end-to-end speech synthesis system. The emotion expressiveness of conventional approach was limited because it utilized only one style representative per each emotion. We overcome the problem by extracting multiple number of representatives per each emotion using a k-means clustering algorithm. Through the results of listening tests, it is proved that the proposed method clearly express each emotion while distinguishing one emotion from others.

• Phonetics and Speech Sciences
• /
• v.13 no.4
• /
• pp.47-53
• /
• 2021

Autoregressive Text-to-Speech (TTS) models suffer from inference instability and slow inference speed. Inference instability occurs when a poorly predicted sample at time step t affects all the subsequent predictions. Slow inference speed arises from a model structure that forces the predicted samples from time steps 1 to t-1 to predict the sample at time step t. In this study, an end-to-end non-autoregressive fast text-to-speech model is suggested as a solution to these problems. The results of this study show that this model's Mean Opinion Score (MOS) is close to that of Tacotron 2 - WaveNet, while this model's inference speed and stability are higher than those of Tacotron 2 - WaveNet. Further, this study aims to offer insight into the improvement of non-autoregressive models.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.453-455
• /
• 2020

본 논문은 시각 장애인을 위한 감정 음성 자막 서비스를 생성하는 종단 간(end-to-end) 감정 음성 합성 시스템(emotional text-to-speech synthesis system, TTS)의 음성 합성 속도를 높이면서도 합성음의 음질을 향상시키는 방법을 제안한다. 기존에 사용했던 전역 스타일 토큰(Global Style Token, GST)을 이용한 감정 음성 합성 방법은 다양한 감정을 표현할 수 있는 장점을 갖고 있으나, 합성음을 생성하는데 필요한 시간이 길고 학습할 데이터의 동적 영역을 효과적으로 처리하지 않으면 합성음에 클리핑(clipping) 현상이 발생하는 등 음질이 저하되는 양상을 보였다. 이를 보안하기 위해 본 논문에서는 새로운 데이터 전처리 과정을 도입하였고 기존의 보코더(vocoder)인 웨이브넷(WaveNet)을 웨이브알엔엔(WaveRNN)으로 대체하여 생성 속도와 음질 측면에서 개선됨을 보였다.

• Journal of the Korea Society of Computer and Information
• /
• v.20 no.5
• /
• pp.53-63
• /
• 2015

Singing voice synthesis is the generation of a song using a computer given its lyrics and musical notes. Hidden Markov models (HMM) have been proved to be the models of choice for text to speech synthesis. HMMs have also been used for singing voice synthesis research, however, a huge database is needed for the training of HMMs for singing voice synthesis. And commercially available singing voice synthesis systems which use the piano roll music notation, needs to adopt the easy to read standard music notation which make it suitable for singing learning applications. To overcome this problem, we use a speech database for training context dependent HMMs, to be used for singing voice synthesis. Pitch and duration control methods have been devised to modify the parameters of the HMMs trained on speech, to be used as the synthesis units for the singing voice. This work describes a singing voice synthesis system which uses a MusicXML based music score editor as the front-end interface for entry of the notes and lyrics to be synthesized and a hidden Markov model based text to speech synthesis system as the back-end synthesizer. A perceptual test shows the feasibility of our proposed system.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.1435-1438
• /
• 2005

In hands-free telephone systems, the received speech signal is fed back to the microphone and constitutes the so-called echo. To cancel the effect of this time-varying echo path, it is necessary to device an adaptive filter between the receiving and the transmitting ends. For a typical FIR realization, the length of the fullband adaptive filter results in high computational complexity and low convergence rate. Consequently, subband adaptive filtering schemes have been proposed to improve the performance. In this work, we use deterministic approach to analyze the relationship between fullband and subband adaptive filtering structures. With block adaptive filtering structure as an intermediate stage, the analysis is divided into two parts. First, to avoid aliasing, it is found that the matrix of block adaptive filters is in the form of pseudocirculant, and the elements of this matrix are the polyphase components of the fullband adaptive filter. Second, to transmit the near-end voice signal faithfully, the analysis and the synthesis filter banks in the subband adaptive filtering structure must form a perfect reconstruction pair. Using polyphase representation, the relationship between the block and the subband adaptive filters is derived.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.359-366
• /
• 2022

In this paper, we present the Korean menu-ordering Sentence Text-to-Speech (TTS) system using conformer-based FastSpeech2. Conformer is the convolution-augmented transformer, which was originally proposed in Speech Recognition. Combining two different structures, the Conformer extracts better local and global features. It comprises two half Feed Forward module at the front and the end, sandwiching the Multi-Head Self-Attention module and Convolution module. We introduce the Conformer in Korean TTS, as we know it works well in Korean Speech Recognition. For comparison between transformer-based TTS model and Conformer-based one, we train FastSpeech2 and Conformer-based FastSpeech2. We collected a phoneme-balanced data set and used this for training our models. This corpus comprises not only general conversation, but also menu-ordering conversation consisting mainly of loanwords. This data set is the solution to the current Korean TTS model's degradation in loanwords. As a result of generating a synthesized sound using ParallelWave Gan, the Conformer-based FastSpeech2 achieved superior performance of MOS 4.04. We confirm that the model performance improved when the same structure was changed from transformer to Conformer in the Korean TTS.