This paper advances phrase break prediction (also known as phrasing) in multi-speaker text-to-speech (TTS) systems. We integrate speaker-specific features by leveraging speaker embeddings to enhance the performance of the phrasing model. We further demonstrate that these speaker embeddings can capture speaker-related characteristics solely from the phrasing task. Besides, we explore the potential of pre-trained speaker embeddings for unseen speakers through a few-shot adaptation method. Furthermore, we pioneer the application of phoneme-level pre-trained language models to this TTS front-end task, which significantly boosts the accuracy of the phrasing model. Our methods are rigorously assessed through both objective and subjective evaluations, demonstrating their effectiveness. • Speaker-conditioned phrasing model improves accuracy in multi-speaker phrasing tasks. • We explore various speaker embeddings in phrasing models. • We apply phoneme-level pre-trained language models to enhance phrasing accuracy. • We propose a speaker adaptation method for few-shot phrasing tasks. • We verify that speaker embeddings learn human-aligned features via phrasing tasks.

Synthetic speech is a useful source for augmentation of automatic speech recognition (ASR) systems, but there is a "sim-to-real" gap between synthetic and real speech that can limit generalization. The natural variability of real speech is essential to the training of robust ASR systems. While synthetic data augmentation can be used to approximate the variability of natural speech, however, not all aspects of variation are equally relevant for augmentation. In this work, we introduce speech re-painting, a method for in-context augmented synthesis, using target training datasets to generate new utterances guided by speech and text on the fly in a zero-shot manner. We evaluate this technique using downstream ASR word error rate (WER) using the VCTK and LibriSpeech datasets. These represent unique speaker and lexical challenges that are addressed by re-painting, realizing a reduction of WER more than 50% in particular settings.

Collecting high-quality studio recordings of audio is challenging, which limits the language coverage of text-to-speech (TTS) systems. This paper proposes a framework for scaling a multilingual TTS model to 100+ languages using found data without supervision. The proposed framework combines speech-text encoder pretraining with unsupervised training using untranscribed speech and unspoken text data sources, thereby leveraging massively multilingual joint speech and text representation learning. Without any transcribed speech in a new language, this TTS model can generate intelligible speech in >30 unseen languages (CER difference of <10% to ground truth). With just 15 minutes of transcribed, found data, we can reduce the intelligibility difference to 1% or less from the ground-truth, and achieve naturalness scores that match the ground-truth in several languages.

This paper proposes a method for extracting a lightweight subset from a text-to-speech (TTS) corpus ensuring synthetic speech quality. In recent years, methods have been proposed for constructing large-scale TTS corpora by collecting diverse data from massive sources such as audiobooks and YouTube. Although these methods have gained significant attention for enhancing the expressive capabilities of TTS systems, they often prioritize collecting vast amounts of data without considering practical constraints like storage capacity and computation time in training, which limits the available data quantity. Consequently, the need arises to efficiently collect data within these volume constraints. To address this, we propose a method for selecting the core subset (known as core-set) from a TTS corpus on the basis of a diversity metric, which measures the degree to which a subset encompasses a wide range. Experimental results demonstrate that our proposed method performs significantly better than the baseline phoneme-balanced data selection across language and corpus size.

Neural text-to-speech (TTS) systems have made significant progress in generating natural synthetic speech. However, neural TTS requires large amounts of paired training data, which limits its applicability to a small number of resource-rich languages. Previous work on low-resource TTS has addressed the data hungriness based on transfer learning from a multilingual model to low-resource languages, but it still relies heavily on the availability of paired data for the target languages. In this paper, we propose a text-inductive language adaptation framework for low-resource TTS to address the cost of collecting the paired data for low-resource languages. To inject textual knowledge during transfer learning, our framework employs a two-stage adaptation scheme that utilizes both text-only and supervised data for the target language. In the text-based adaptation stage, we update the language-aware embedding layer with a masked language model objective using text-only data for the target language. In the supervised adaptation stage, the entire TTS model is updated using paired data for the target language. We also propose a graphone-based multilingual training method that jointly uses graphemes and International Phonetic Alphabet symbols (referred to as graphones) for resource-rich languages, while using only graphemes for low-resource languages. This approach facilitates the transfer of pronunciation knowledge from resource-rich to low-resource languages. Through extensive evaluations, we demonstrate that 1) our framework with text-based adaptation outperforms the previous supervised transfer learning approach, 2) the proposed graphone-based training method further improves the performance of both multilingual TTS and low-resource language adaptation. With only 5 minutes of paired data for fine-tuning, our method achieved highly intelligible synthetic speech with the character error rates of around 6 % for a target language.

In this study, we introduce YODAS (YouTube-Oriented Dataset for Audio and Speech), a large-scale, multilingual dataset comprising currently over 500k hours of speech data in more than 100 languages, sourced from both labeled and unlabeled YouTube speech datasets. The labeled subsets, including manual or automatic subtitles, facilitate supervised model training. Conversely, the unlabeled subsets are apt for self-supervised learning applications. YODAS is distinctive as the first publicly available dataset of its scale, and it will be distributed under a Creative Commons license. We introduce the collection methodology utilized for YODAS, which contributes to the large-scale speech dataset construction. Subsequently, we provide a comprehensive analysis of speech, text contained within the dataset. Finally, we describe the speech recognition baselines over the top-15 languages.

While human evaluation is the most reliable metric for evaluating speech generation systems, it is generally costly and time-consuming. Previous studies on automatic speech quality assessment address the problem by predicting human evaluation scores with machine learning models. However, they rely on supervised learning and thus suffer from high annotation costs and domain-shift problems. We propose SpeechLMScore, an unsupervised metric to evaluate generated speech using a speech language model. SpeechLMScore computes the average log-probability of a speech signal by mapping it into discrete tokens and measures the average probability of generating the sequence of tokens. Therefore, it does not require human annotation and is a highly scalable framework. Evaluation results demonstrate that the proposed metric shows a promising correlation with human evaluation scores on different speech generation tasks including voice conversion, text-to-speech, and speech enhancement.

This paper proposes Virtuoso, a massively multilingual speech–text joint semi-supervised learning framework for text-to-speech synthesis (TTS) models. Existing multilingual TTS typically supports tens of languages, which are a small fraction of the thousands of languages in the world. One difficulty to scale multilingual TTS to hundreds of languages is collecting high-quality speech–text paired data in low-resource languages. This study extends Maestro, a speech–text joint pretraining framework for automatic speech recognition (ASR), to speech generation tasks. To train a TTS model from various types of speech and text data, different training schemes are designed to handle supervised (paired TTS and ASR data) and unsupervised (untranscribed speech and unspoken text) datasets. Experimental evaluation shows that 1) multilingual TTS models trained on Virtuoso can achieve significantly better naturalness and intelligibility than baseline ones in seen languages, and 2) they can synthesize reasonably intelligible and naturally sounding speech for unseen languages where no high-quality paired TTS data is available.

Pause insertion, also known as phrase break prediction and phrasing, is an essential part of TTS systems because proper pauses with natural duration significantly enhance the rhythm and intelligibility of synthetic speech. However, conventional phrasing models ignore various speakers’ different styles of inserting silent pauses, which can degrade the performance of the model trained on a multi-speaker speech corpus. To this end, we propose more powerful pause insertion frameworks based on a pre-trained language model. Our approach uses bidirectional encoder representations from transformers (BERT) pre-trained on a large-scale text corpus, injecting speaker embeddings to capture various speaker characteristics. We also leverage duration-aware pause insertion for more natural multi-speaker TTS. We develop and evaluate two types of models. The first improves conventional phrasing models on the position prediction of respiratory pauses (RPs), i.e., silent pauses at word transitions without punctuation. It performs speaker-conditioned RP prediction considering contextual information and is used to demonstrate the effect of speaker information on the prediction. The second model is further designed for phoneme-based TTS models and performs duration-aware pause insertion, predicting both RPs and punctuation-indicated pauses (PIPs) that are categorized by duration. The evaluation results show that our models improve the precision and recall of pause insertion and the rhythm of synthetic speech.

While neural text-to-speech (TTS) has achieved human-like natural synthetic speech, multilingual TTS systems are limited to resource-rich languages due to the need for paired text and studio-quality audio data. This paper proposes a method for zero-shot multilingual TTS using text-only data for the target language. The use of text-only data allows the development of TTS systems for low-resource languages for which only textual resources are available, making TTS accessible to thousands of languages. Inspired by the strong cross-lingual transferability of multilingual language models, our framework first performs masked language model pretraining with multilingual text-only data. Then we train this model with a paired data in a supervised manner, while freezing a language-aware embedding layer. This allows inference even for languages not included in the paired data but present in the text-only data. Evaluation results demonstrate highly intelligible zero-shot TTS with a character error rate of less than 12% for an unseen language.

We present a self-supervised speech restoration method without paired speech corpora.Because the previous general speech restoration method uses artificial paired data created by applying various distortions to high-quality speech corpora, it cannot sufficiently represent acoustic distortions of real data, limiting the applicability.Our model consists of analysis, synthesis, and channel modules that simulate the recording process of degraded speech and is trained with real degraded speech data in a self-supervised manner.The analysis module extracts distortionless speech features and distortion features from degraded speech, while the synthesis module synthesizes the restored speech waveform, and the channel module adds distortions to the speech waveform.Our model also enables audio effect transfer, in which only acoustic distortions are extracted from degraded speech and added to arbitrary high-quality audio.Experimental evaluations with both simulated and real data show that our method achieves significantly higher-quality speech restoration than the previous supervised method, suggesting its applicability to real degraded speech materials.

Most text-to-speech (TTS) methods use high-quality speech corpora recorded in a well-designed environment, incurring a high cost for data collection.To solve this problem, existing noise-robust TTS methods are intended to use noisy speech corpora as training data.However, they only address either timeinvariant or time-variant noises.We propose a degradationrobust TTS method, which can be trained on speech corpora that contain both additive noises and environmental distortions.It jointly represents the time-variant additive noises with a framelevel encoder and the time-invariant environmental distortions with an utterance-level encoder.We also propose a regularization method to attain clean environmental embedding that is disentangled from the utterance-dependent information such as linguistic contents and speaker characteristics.Evaluation results show that our method achieved significantly higher-quality synthetic speech than previous methods in the condition including both additive noise and reverberation.

We present the UTokyo-SaruLab mean opinion score (MOS) prediction system submitted to VoiceMOS Challenge 2022.The challenge is to predict the MOS values of speech samples collected from previous Blizzard Challenges and Voice Conversion Challenges for two tracks: a main track for in-domain prediction and an out-of-domain (OOD) track for which there is less labeled data from different listening tests.Our system is based on ensemble learning of strong and weak learners.Strong learners incorporate several improvements to the previous finetuning models of self-supervised learning (SSL) models, while weak learners use basic machine-learning methods to predict scores from SSL features.In the Challenge, our system had the highest score on several metrics for both the main and OOD tracks.In addition, we conducted ablation studies to investigate the effectiveness of our proposed methods.

This article presents SSR7000, a corpus of synchronized ultrasound tongue and lip images designed for end-to-end silent speech recognition (SSR). Although neural end-to-end models are successfully updating the state-of-the-art technology in the field of automatic speech recognition, SSR research based on ultrasound tongue imaging has still not evolved past cascaded DNN-HMM models due to the absence of a large dataset. In this study, we constructed a large dataset, namely SSR7000, to exploit the performance of the end-to-end models. The SSR7000 dataset contains ultrasound tongue and lip images of 7484 utterances by a single speaker. It contains more utterances per person than any other SSR corpus based on ultrasound imaging. We also describe preprocessing techniques to tackle data variances that are inevitable when collecting a large dataset and present benchmark results using an end-to-end model. The SSR7000 corpus is publicly available under the CC BY-NC 4.0 license.

In this paper, we propose a method to generate personalized filled pauses (FPs) with group-wise prediction models. Compared with fluent text generation, disfluent text generation has not been widely explored. To generate more human-like texts, we addressed disfluent text generation. The usage of disfluency, such as FPs, rephrases, and word fragments, differs from speaker to speaker, and thus, the generation of personalized FPs is required. However, it is difficult to predict them because of the sparsity of position and the frequency difference between more and less frequently used FPs. Moreover, it is sometimes difficult to adapt FP prediction models to each speaker because of the large variation of the tendency within each speaker. To address these issues, we propose a method to build group-dependent prediction models by grouping speakers on the basis of their tendency to use FPs. This method does not require a large amount of data and time to train each speaker model. We further introduce a loss function and a word embedding model suitable for FP prediction. Our experimental results demonstrate that group-dependent models can predict FPs with higher scores than a non-personalized one and the introduced loss function and word embedding model improve the prediction performance.

This paper proposes a visual-text to speech (vTTS) method, a method for synthesizing speech directly from visual text (i.e., text as an image). vTTS can use visual features in visual text that should be important for speech synthesis such as emphasis and radicals (components in Chinese characters), but they are not available in conventional TTS using discrete symbols as the input. The proposed vTTS method extracts visual features with a convolutional neural network and then generates acoustic features with a non-autoregressive model in an end-to-end manner. Experimental results show that 1) the vTTS method is capable of generating speech with naturalness comparable to or better than a conventional TTS, 2) it can transfer emphasis and emotion attributes in visual text to speech without additional labels and architectures, and 3) it can synthesize more natural and intelligible speech from unseen and rare characters than conventional TTS.

Incremental text-to-speech (TTS) synthesis generates utterances in small linguistic units for the sake of real-time and low-latency applications. We previously proposed an incremental TTS method that leverages a large pre-trained language model to take unobserved future context into account without waiting for the subsequent segment. Although this method achieves comparable speech quality to that of a method that waits for the future context, it entails a huge amount of processing for sampling from the language model at each time step. In this paper, we propose an incremental TTS method that directly predicts the unobserved future context with a lightweight model, instead of sampling words from the large-scale language model. We perform knowledge distillation from a GPT2-based context prediction network into a simple recurrent model by minimizing a teacher-student loss defined between the context embedding vectors of those models. Experimental results show that the proposed method requires about ten times less inference time to achieve comparable synthetic speech quality to that of our previous method, and it can perform incremental synthesis much faster than the average speaking speed of human English speakers, demonstrating the availability of our method to real-time applications.

This letter presents an incremental text-to-speech (TTS) method that performs synthesis in small linguistic units while maintaining the naturalness of output speech. Incremental TTS is generally subject to a trade-off between latency and synthetic speech quality. It is challenging to produce high-quality speech with a low-latency setup that does not make much use of an unobserved future sentence (hereafter, “lookahead”). To resolve this issue, we propose an incremental TTS method that uses a pseudo lookahead generated with a language model to take the future contextual information into account without increasing latency. Our method can be regarded as imitating a human's incremental reading and uses pretrained GPT2, which accounts for the large-scale linguistic knowledge, for the lookahead generation. Evaluation results show that our method 1) achieves higher speech quality than the method taking only observed information into account and 2) achieves a speech quality equivalent to waiting for the future context observation.

In this paper, we propose computationally efficient and high-quality methods for statistical voice conversion (VC) with direct waveform modification based on spectral differentials. The conventional method with a minimum-phase filter achieves high-quality conversion but requires heavy computation in filtering. This is because the minimum phase using a fixed lifter of the Hilbert transform often results in a long-tap filter. One of our methods is a data-driven method for lifter training. Since this method takes filter truncation into account in training, it can shorten the tap length of the filter while preserving conversion accuracy. Our other method is sub-band processing for extending the conventional method from narrow-band (16 kHz) to full-band (48 kHz) VC, which can convert a full-band waveform with higher converted-speech quality. Experimental results indicate that 1) the proposed lifter-training method for narrow-band VC can shorten the tap length to 1/16 without degrading the converted-speech quality and 2) the proposed sub-band-processing method for full-band VC can improve the converted-speech quality than the conventional method.