Kenichi FUJITA Atsushi ANDO Yusuke IJIMA
This paper proposes a speech rhythm-based method for speaker embeddings to model phoneme duration using a few utterances by the target speaker. Speech rhythm is one of the essential factors among speaker characteristics, along with acoustic features such as F0, for reproducing individual utterances in speech synthesis. A novel feature of the proposed method is the rhythm-based embeddings extracted from phonemes and their durations, which are known to be related to speaking rhythm. They are extracted with a speaker identification model similar to the conventional spectral feature-based one. We conducted three experiments, speaker embeddings generation, speech synthesis with generated embeddings, and embedding space analysis, to evaluate the performance. The proposed method demonstrated a moderate speaker identification performance (15.2% EER), even with only phonemes and their duration information. The objective and subjective evaluation results demonstrated that the proposed method can synthesize speech with speech rhythm closer to the target speaker than the conventional method. We also visualized the embeddings to evaluate the relationship between the distance of the embeddings and the perceptual similarity. The visualization of the embedding space and the relation analysis between the closeness indicated that the distribution of embeddings reflects the subjective and objective similarity.
Kento MATSUMOTO Sunao HARA Masanobu ABE
In this paper, we propose a new algorithm to generate Speech-like Emotional Sound (SES). Emotional expressions may be the most important factor in human communication, and speech is one of the most useful means of expressing emotions. Although speech generally conveys both emotional and linguistic information, we have undertaken the challenge of generating sounds that convey emotional information alone. We call the generated sounds “speech-like,” because the sounds do not contain any linguistic information. SES can provide another way to generate emotional response in human-computer interaction systems. To generate “speech-like” sound, we propose employing WaveNet as a sound generator conditioned only by emotional IDs. This concept is quite different from the WaveNet Vocoder, which synthesizes speech using spectrum information as an auxiliary feature. The biggest advantage of our approach is that it reduces the amount of emotional speech data necessary for training by focusing on non-linguistic information. The proposed algorithm consists of two steps. In the first step, to generate a variety of spectrum patterns that resemble human speech as closely as possible, WaveNet is trained with auxiliary mel-spectrum parameters and Emotion ID using a large amount of neutral speech. In the second step, to generate emotional expressions, WaveNet is retrained with auxiliary Emotion ID only using a small amount of emotional speech. Experimental results reveal the following: (1) the two-step training is necessary to generate the SES with high quality, and (2) it is important that the training use a large neutral speech database and spectrum information in the first step to improve the emotional expression and naturalness of SES.
Bima PRIHASTO Tzu-Chiang TAI Pao-Chi CHANG Jia-Ching WANG
The recurrent neural network (RNN) has been used in audio and speech processing, such as language translation and speech recognition. Although RNN-based architecture can be applied to speech synthesis, the long computing time is still the primary concern. This research proposes a fast gated recurrent neural network, a fast RNN-based architecture, for speech synthesis based on the minimal gated unit (MGU). Our architecture removes the unit state history from some equations in MGU. Our MGU-based architecture is about twice faster, with equally good sound quality than the other MGU-based architectures.
Kiyoshi KURIHARA Nobumasa SEIYAMA Tadashi KUMANO
This paper describes a method to control prosodic features using phonetic and prosodic symbols as input of attention-based sequence-to-sequence (seq2seq) acoustic modeling (AM) for neural text-to-speech (TTS). The method involves inserting a sequence of prosodic symbols between phonetic symbols that are then used to reproduce prosodic acoustic features, i.e. accents, pauses, accent breaks, and sentence endings, in several seq2seq AM methods. The proposed phonetic and prosodic labels have simple descriptions and a low production cost. By contrast, the labels of conventional statistical parametric speech synthesis methods are complicated, and the cost of time alignments such as aligning the boundaries of phonemes is high. The proposed method does not need the boundary positions of phonemes. We propose an automatic conversion method for conventional labels and show how to automatically reproduce pitch accents and phonemes. The results of objective and subjective evaluations show the effectiveness of our method.
Junya KOGUCHI Shinnosuke TAKAMICHI Masanori MORISE Hiroshi SARUWATARI Shigeki SAGAYAMA
We propose a speech analysis-synthesis and deep neural network (DNN)-based text-to-speech (TTS) synthesis framework using Gaussian mixture model (GMM)-based approximation of full-band spectral envelopes. GMMs have excellent properties as acoustic features in statistic parametric speech synthesis. Each Gaussian function of a GMM fits the local resonance of the spectrum. The GMM retains the fine spectral envelope and achieve high controllability of the structure. However, since conventional speech analysis methods (i.e., GMM parameter estimation) have been formulated for a narrow-band speech, they degrade the quality of synthetic speech. Moreover, a DNN-based TTS synthesis method using GMM-based approximation has not been formulated in spite of its excellent expressive ability. Therefore, we employ peak-picking-based initialization for full-band speech analysis to provide better initialization for iterative estimation of the GMM parameters. We introduce not only prediction error of GMM parameters but also reconstruction error of the spectral envelopes as objective criteria for training DNN. Furthermore, we propose a method for multi-task learning based on minimizing these errors simultaneously. We also propose a post-filter based on variance scaling of the GMM for our framework to enhance synthetic speech. Experimental results from evaluating our framework indicated that 1) the initialization method of our framework outperformed the conventional one in the quality of analysis-synthesized speech; 2) introducing the reconstruction error in DNN training significantly improved the synthetic speech; 3) our variance-scaling-based post-filter further improved the synthetic speech.
Mohammed Salah AL-RADHI Tamás Gábor CSAPÓ Géza NÉMETH
In this article, we propose a method called “continuous noise masking (cNM)” that allows eliminating residual buzziness in a continuous vocoder, i.e. of which all parameters are continuous and offers a simple and flexible speech analysis and synthesis system. Traditional parametric vocoders generally show a perceptible deterioration in the quality of the synthesized speech due to different processing algorithms. Furthermore, an inaccurate noise resynthesis (e.g. in breathiness or hoarseness) is also considered to be one of the main underlying causes of performance degradation, leading to noisy transients and temporal discontinuity in the synthesized speech. To overcome these issues, a new cNM is developed based on the phase distortion deviation in order to reduce the perceptual effect of the residual noise, allowing a proper reconstruction of noise characteristics, and model better the creaky voice segments that may happen in natural speech. To this end, the cNM is designed to keep only voice components under a condition of the cNM threshold while discarding others. We evaluate the proposed approach and compare with state-of-the-art vocoders using objective and subjective listening tests. Experimental results show that the proposed method can reduce the effect of residual noise and can reach the quality of other sophisticated approaches like STRAIGHT and log domain pulse model (PML).
The effectiveness of model adaptation in dialogue speech synthesis is explored. The proposed adaptation method is based on a conversion from a base model learned with a large dataset into a target, dialogue-style speech model. The proposed method is shown to improve the intelligibility of synthesized dialogue speech, while maintaining the speaking style of dialogue.
Conventional approaches to statistical parametric speech synthesis use context-dependent hidden Markov models (HMMs) clustered using decision trees to generate speech parameters from linguistic features. However, decision trees are not always appropriate to model complex context dependencies of linguistic features efficiently. An alternative scheme that replaces decision trees with deep neural networks (DNNs) was presented as a possible way to overcome the difficulty. By training the network to represent high-dimensional feedforward dependencies from linguistic features to acoustic features, DNN-based speech synthesis systems convert a text into a speech. To improved the naturalness of the synthesized speech, this paper presents a novel pre-training method for DNN-based statistical parametric speech synthesis systems. In our method, a deep relational model (DRM), which represents a joint probability of two visible variables, is applied to describe the joint distribution of acoustic and linguistic features. As with DNNs, a DRM consists several hidden layers and two visible layers. Although DNNs represent feedforward dependencies from one visible variables (inputs) to other visible variables (outputs), a DRM has an ability to represent the bidirectional dependencies between two visible variables. During the maximum-likelihood (ML) -based training, the model optimizes its parameters (connection weights between two adjacent layers, and biases) of a deep architecture considering the bidirectional conversion between 1) acoustic features given linguistic features, and 2) linguistic features given acoustic features generated from itself. Owing to considering whether the generated acoustic features are recognizable, our method can obtain reasonable parameters for speech synthesis. Experimental results in a speech synthesis task show that pre-trained DNN-based systems using our proposed method outperformed randomly-initialized DNN-based systems, especially when the amount of training data is limited. Additionally, speaker-dependent speech recognition experimental results also show that our method outperformed DNN-based systems, by setting the initial parameters of our method are the same as that in the synthesis experiments.
Daiki SEKIZAWA Shinnosuke TAKAMICHI Hiroshi SARUWATARI
This article proposes a prosody correction method based on partial model adaptation for Chinese-accented Japanese hidden Markov model (HMM)-based text-to-speech synthesis. Although text-to-speech synthesis built from non-native speech accurately reproduces the speaker's individuality in synthetic speech, the naturalness of the synthetic speech is strongly degraded. In the proposed model, to improve the naturalness while preserving the speaker individuality of Chinese-accented Japanese text-to-speech synthesis, we partially utilize HMM parameters of native Japanese speech to synthesize prosody-corrected synthetic speech. Results of an experimental evaluation demonstrate that duration and F0 correction are significantly effective for improving naturalness.
Nobukatsu HOJO Yusuke IJIMA Hideyuki MIZUNO
Deep neural network (DNN)-based speech synthesis can produce more natural synthesized speech than the conventional HMM-based speech synthesis. However, it is not revealed whether the synthesized speech quality can be improved by utilizing a multi-speaker speech corpus. To address this problem, this paper proposes DNN-based speech synthesis using speaker codes as a method to improve the performance of the conventional speaker dependent DNN-based method. In order to model speaker variation in the DNN, the augmented feature (speaker codes) is fed to the hidden layer(s) of the conventional DNN. This paper investigates the effectiveness of introducing speaker codes to DNN acoustic models for speech synthesis for two tasks: multi-speaker modeling and speaker adaptation. For the multi-speaker modeling task, the method we propose trains connection weights of the whole DNN using a multi-speaker speech corpus. When performing multi-speaker synthesis, the speaker code corresponding to the selected target speaker is fed to the DNN to generate the speaker's voice. When performing speaker adaptation, a set of connection weights of the multi-speaker model is re-estimated to generate a new target speaker's voice. We investigated the relationship between the prediction performance and architecture of the DNNs through objective measurements. Objective evaluation experiments revealed that the proposed model outperformed conventional methods (HMMs, speaker dependent DNNs and multi-speaker DNNs based on a shared hidden layer structure). Subjective evaluation experimental results showed that the proposed model again outperformed the conventional methods (HMMs, speaker dependent DNNs), especially when using a small number of target speaker utterances.
Yuki SAITO Shinnosuke TAKAMICHI Hiroshi SARUWATARI
This paper proposes Deep Neural Network (DNN)-based Voice Conversion (VC) using input-to-output highway networks. VC is a speech synthesis technique that converts input features into output speech parameters, and DNN-based acoustic models for VC are used to estimate the output speech parameters from the input speech parameters. Given that the input and output are often in the same domain (e.g., cepstrum) in VC, this paper proposes a VC using highway networks connected from the input to output. The acoustic models predict the weighted spectral differentials between the input and output spectral parameters. The architecture not only alleviates over-smoothing effects that degrade speech quality, but also effectively represents the characteristics of spectral parameters. The experimental results demonstrate that the proposed architecture outperforms Feed-Forward neural networks in terms of the speech quality and speaker individuality of the converted speech.
Shigeki MATSUDA Teruaki HAYASHI Yutaka ASHIKARI Yoshinori SHIGA Hidenori KASHIOKA Keiji YASUDA Hideo OKUMA Masao UCHIYAMA Eiichiro SUMITA Hisashi KAWAI Satoshi NAKAMURA
This study introduces large-scale field experiments of VoiceTra, which is the world's first speech-to-speech multilingual translation application for smart phones. In the study, approximately 10 million input utterances were collected since the experiments commenced. The usage of collected data was analyzed and discussed. The study has several important contributions. First, it explains system configuration, communication protocol between clients and servers, and details of multilingual automatic speech recognition, multilingual machine translation, and multilingual speech synthesis subsystems. Second, it demonstrates the effects of mid-term system updates using collected data to improve an acoustic model, a language model, and a dictionary. Third, it analyzes system usage.
Nobuaki MINEMATSU Ibuki NAKAMURA Masayuki SUZUKI Hiroko HIRANO Chieko NAKAGAWA Noriko NAKAMURA Yukinori TAGAWA Keikichi HIROSE Hiroya HASHIMOTO
This paper develops an online and freely available framework to aid teaching and learning the prosodic control of Tokyo Japanese: how to generate its adequate word accent and phrase intonation. This framework is called OJAD (Online Japanese Accent Dictionary) [1] and it provides three features. 1) Visual, auditory, systematic, and comprehensive illustration of patterns of accent change (accent sandhi) of verbs and adjectives. Here only the changes caused by twelve fundamental conjugations are focused upon. 2) Visual illustration of the accent pattern of a given verbal expression, which is a combination of a verb and its postpositional auxiliary words. 3) Visual illustration of the pitch pattern of any given sentence and the expected positions of accent nuclei in the sentence. The third feature is technically implemented by using an accent change prediction module that we developed for Japanese Text-To-Speech (TTS) synthesis [2],[3]. Experiments show that accent nucleus assignment to given texts by the proposed framework is much more accurate than that by native speakers. Subjective assessment and objective assessment done by teachers and learners show extremely high pedagogical effectiveness of the developed framework.
Yuji OSHIMA Shinnosuke TAKAMICHI Tomoki TODA Graham NEUBIG Sakriani SAKTI Satoshi NAKAMURA
This paper presents a novel non-native speech synthesis technique that preserves the individuality of a non-native speaker. Cross-lingual speech synthesis based on voice conversion or Hidden Markov Model (HMM)-based speech synthesis is a technique to synthesize foreign language speech using a target speaker's natural speech uttered in his/her mother tongue. Although the technique holds promise to improve a wide variety of applications, it tends to cause degradation of target speaker's individuality in synthetic speech compared to intra-lingual speech synthesis. This paper proposes a new approach to speech synthesis that preserves speaker individuality by using non-native speech spoken by the target speaker. Although the use of non-native speech makes it possible to preserve the speaker individuality in the synthesized target speech, naturalness is significantly degraded as the synthesized speech waveform is directly affected by unnatural prosody and pronunciation often caused by differences in the linguistic systems of the source and target languages. To improve naturalness while preserving speaker individuality, we propose (1) a prosody correction method based on model adaptation, and (2) a phonetic correction method based on spectrum replacement for unvoiced consonants. The experimental results using English speech uttered by native Japanese speakers demonstrate that (1) the proposed methods are capable of significantly improving naturalness while preserving the speaker individuality in synthetic speech, and (2) the proposed methods also improve intelligibility as confirmed by a dictation test.
Xin WANG Shinji TAKAKI Junichi YAMAGISHI
Building high-quality text-to-speech (TTS) systems without expert knowledge of the target language and/or time-consuming manual annotation of speech and text data is an important yet challenging research topic. In this kind of TTS system, it is vital to find representation of the input text that is both effective and easy to acquire. Recently, the continuous representation of raw word inputs, called “word embedding”, has been successfully used in various natural language processing tasks. It has also been used as the additional or alternative linguistic input features to a neural-network-based acoustic model for TTS systems. In this paper, we further investigate the use of this embedding technique to represent phonemes, syllables and phrases for the acoustic model based on the recurrent and feed-forward neural network. Results of the experiments show that most of these continuous representations cannot significantly improve the system's performance when they are fed into the acoustic model either as additional component or as a replacement of the conventional prosodic context. However, subjective evaluation shows that the continuous representation of phrases can achieve significant improvement when it is combined with the prosodic context as input to the acoustic model based on the feed-forward neural network.
Shinnosuke TAKAMICHI Tomoki TODA Graham NEUBIG Sakriani SAKTI Satoshi NAKAMURA
This paper presents a novel statistical sample-based approach for Gaussian Mixture Model (GMM)-based Voice Conversion (VC). Although GMM-based VC has the promising flexibility of model adaptation, quality in converted speech is significantly worse than that of natural speech. This paper addresses the problem of inaccurate modeling, which is one of the main reasons causing the quality degradation. Recently, we have proposed statistical sample-based speech synthesis using rich context models for high-quality and flexible Hidden Markov Model (HMM)-based Text-To-Speech (TTS) synthesis. This method makes it possible not only to produce high-quality speech by introducing ideas from unit selection synthesis, but also to preserve flexibility of the original HMM-based TTS. In this paper, we apply this idea to GMM-based VC. The rich context models are first trained for individual joint speech feature vectors, and then we gather them mixture by mixture to form a Rich context-GMM (R-GMM). In conversion, an iterative generation algorithm using R-GMMs is used to convert speech parameters, after initialization using over-trained probability distributions. Because the proposed method utilizes individual speech features, and its formulation is the same as that of conventional GMM-based VC, it makes it possible to produce high-quality speech while keeping flexibility of the original GMM-based VC. The experimental results demonstrate that the proposed method yields significant improvements in term of speech quality and speaker individuality in converted speech.
Masanori MORISE Fumiya YOKOMORI Kenji OZAWA
A vocoder-based speech synthesis system, named WORLD, was developed in an effort to improve the sound quality of real-time applications using speech. Speech analysis, manipulation, and synthesis on the basis of vocoders are used in various kinds of speech research. Although several high-quality speech synthesis systems have been developed, real-time processing has been difficult with them because of their high computational costs. This new speech synthesis system has not only sound quality but also quick processing. It consists of three analysis algorithms and one synthesis algorithm proposed in our previous research. The effectiveness of the system was evaluated by comparing its output with against natural speech including consonants. Its processing speed was also compared with those of conventional systems. The results showed that WORLD was superior to the other systems in terms of both sound quality and processing speed. In particular, it was over ten times faster than the conventional systems, and the real time factor (RTF) indicated that it was fast enough for real-time processing.
Duy Khanh NINH Yoichi YAMASHITA
A conventional HMM-based speech synthesis system for Hanoi Vietnamese often suffers from hoarse quality due to incomplete F0 parameterization of glottalized tones. Since estimating F0 from glottalized waveform is rather problematic for usual F0 extractors, we propose a pitch marking algorithm where pitch marks are propagated from regular regions of a speech signal to glottalized ones, from which complete F0 contours for the glottalized tones are derived. The proposed F0 parameterization scheme was confirmed to significantly reduce the hoarseness whilst slightly improving the tone naturalness of synthetic speech by both objective and listening tests. The pitch marking algorithm works as a refinement step based on the results of an F0 extractor. Therefore, the proposed scheme can be combined with any F0 extractor.
Kazuhiro NAKAMURA Kei HASHIMOTO Yoshihiko NANKAKU Keiichi TOKUDA
This paper proposes a novel approach for integrating spectral feature extraction and acoustic modeling in hidden Markov model (HMM) based speech synthesis. The statistical modeling process of speech waveforms is typically divided into two component modules: the frame-by-frame feature extraction module and the acoustic modeling module. In the feature extraction module, the statistical mel-cepstral analysis technique has been used and the objective function is the likelihood of mel-cepstral coefficients for given speech waveforms. In the acoustic modeling module, the objective function is the likelihood of model parameters for given mel-cepstral coefficients. It is important to improve the performance of each component module for achieving higher quality synthesized speech. However, the final objective of speech synthesis systems is to generate natural speech waveforms from given texts, and the improvement of each component module does not always lead to the improvement of the quality of synthesized speech. Therefore, ideally all objective functions should be optimized based on an integrated criterion which well represents subjective speech quality of human perception. In this paper, we propose an approach to model speech waveforms directly and optimize the final objective function. Experimental results show that the proposed method outperformed the conventional methods in objective and subjective measures.
Chen-Yu YANG Zhen-Hua LING Li-Rong DAI
In this paper, an automatic and unsupervised method using context-dependent hidden Markov models (CD-HMMs) is proposed for the prosodic labeling of speech synthesis databases. This method consists of three main steps, i.e., initialization, model training and prosodic labeling. The initial prosodic labels are obtained by unsupervised clustering using the acoustic features designed according to the characteristics of the prosodic descriptor to be labeled. Then, CD-HMMs of the spectral parameters, F0s and phone durations are estimated by a means similar to the HMM-based parametric speech synthesis using the initial prosodic labels. These labels are further updated by Viterbi decoding under the maximum likelihood criterion given the acoustic feature sequences and the trained CD-HMMs. The model training and prosodic labeling procedures are conducted iteratively until convergence. The performance of the proposed method is evaluated on Mandarin speech synthesis databases and two prosodic descriptors are investigated, i.e., the prosodic phrase boundary and the emphasis expression. In our implementation, the prosodic phrase boundary labels are initialized by clustering the durations of the pauses between every two consecutive prosodic words, and the emphasis expression labels are initialized by examining the differences between the original and the synthetic F0 trajectories. Experimental results show that the proposed method is able to label the prosodic phrase boundary positions much more accurately than the text-analysis-based method without requiring any manually labeled training data. The unit selection speech synthesis system constructed using the prosodic phrase boundary labels generated by our proposed method achieves similar performance to that using the manual labels. Furthermore, the unit selection speech synthesis system constructed using the emphasis expression labels generated by our proposed method can convey the emphasis information effectively while maintaining the naturalness of synthetic speech.