We propose a lightweight neural front-end framework for on-device speech generation and highlight its benefits towards low-resource language scaling. While data-driven models have shown potential in front-end literature, especially since they can enable fast language expansion, they are often extremely large and of high latency. There is limited work focusing on their usability in real-time settings, and none for on-device TTS applications. At the cost of small performances trade-offs, we build lightweight neural Grapheme-to-Phoneme and verbalization models which achieve, on average across three languages, a p90 latency reduction of 95.98% per token on single-threaded [email protected], with respect to a traditional transformer-based baseline, while having 99.26% less parameters. Additionally, leveraging pre-trained teacher models to bootstrap lightweight students, we enable low-resource language scaling on both Grapheme-to-Phoneme conversion and verbalization.

State-of-the-art text-to-speech (TTS) systems require several hours of recorded speech data to generate high-quality synthetic speech. When using reduced amounts of training data, standard TTS models suffer from speech quality and intelligibility degradations, making training low-resource TTS systems problematic. In this paper, we propose a novel extremely low-resource TTS method called Voice Filter that uses as little as one minute of speech from a target speaker. It uses voice conversion (VC) as a post-processing module appended to a pre-existing high-quality TTS system and marks a conceptual shift in the existing TTS paradigm, framing the few-shot TTS problem as a VC task. Furthermore, we propose to use a duration-controllable TTS system to create a parallel speech corpus to facilitate the VC task. Results show that the Voice Filter outperforms state-of-the-art few-shot speech synthesis techniques in terms of objective and subjective metrics on one minute of speech on a diverse set of voices, while being competitive against a TTS model built on 30 times more data.1

We address the problem of cross-speaker style transfer for text-to-speech (TTS) using data augmentation via voice conversion. We assume to have a corpus of neutral non-expressive data from a target speaker and supporting conversational expressive data from different speakers. Our goal is to build a TTS system that is expressive, while retaining the target speaker’s identity. The proposed approach relies on voice conversion to first generate high-quality data from the set of supporting expressive speakers. The voice converted data is then pooled with natural data from the target speaker and used to train a single-speaker multi-style TTS system. We provide evidence that this approach is efficient, flexible, and scalable. The method is evaluated using one or more supporting speakers, as well as a variable amount of supporting data. We further provide evidence that this approach allows some controllability of speaking style, when using multiple supporting speakers. We conclude by scaling our proposed technology to a set of 14 speakers across 7 languages. Results indicate that our technology consistently improves synthetic samples in terms of style similarity, while retaining the target speaker’s identity.

The availability of data in expressive styles across languages is limited, and recording sessions are costly and time consuming.To overcome these issues, we demonstrate how to build low-resource, neural text-to-speech (TTS) voices with only 1 hour of conversational speech, when no other conversational data are available in the same language.Assuming the availability of non-expressive speech data in that language, we propose a 3-step technology: 1) we train an F0-conditioned voice conversion (VC) model as data augmentation technique; 2) we train an F0 predictor to control the conversational flavour of the voice-converted synthetic data; 3) we train a TTS system that consumes the augmented data.We prove that our technology enables F0 controllability, is scalable across speakers and languages and is competitive in terms of naturalness over a state-of-the-art baseline model, another augmented method which does not make use of F0 information.

Automatically predicting the outcome of subjective listening tests is a challenging task. Ratings may vary from person to person even if preferences are consistent across listeners. While previous work has focused on predicting listeners' ratings (mean opinion scores) of individual stimuli, we focus on the simpler task of predicting subjective preference given two speech stimuli for the same text. We propose a model based on anti-symmetric twin neural networks, trained on pairs of waveforms and their corresponding preference scores. We explore both attention and recurrent neural nets to account for the fact that stimuli in a pair are not time aligned. To obtain a large training set we convert listeners' ratings from MUSHRA tests to values that reflect how often one stimulus in the pair was rated higher than the other. Specifically, we evaluate performance on data obtained from twelve MUSHRA evaluations conducted over five years, containing different TTS systems, built from data of different speakers. Our results compare favourably to a state-of-the-art model trained to predict MOS scores.

We investigate multi-speaker speech recognition from ultrasound images of the tongue and video images of the lips. We train our systems on imaging data from modal speech, and evaluate on matched test sets of two speaking modes: silent and modal speech. We observe that silent speech recognition from imaging data underperforms compared to modal speech recognition, likely due to a speaking-mode mismatch between training and testing. We improve silent speech recognition performance using techniques that address the domain mismatch, such as fMLLR and unsupervised model adaptation. We also analyse the properties of silent and modal speech in terms of utterance duration and the size of the articulatory space. To estimate the articulatory space, we compute the convex hull of tongue splines, extracted from ultrasound tongue images. Overall, we observe that the duration of silent speech is longer than that of modal speech, and that silent speech covers a smaller articulatory space than modal speech. Although these two properties are statistically significant across speaking modes, they do not directly correlate with word error rates from speech recognition.

We present the Tongue and Lips corpus (TaL), a multi-speaker corpus of audio, ultrasound tongue imaging, and lip videos. TaL consists of two parts: TaL1 is a set of six recording sessions of one professional voice talent, a male native speaker of English; TaL80 is a set of recording sessions of 81 native speakers of English without voice talent experience. Overall, the corpus contains 24 hours of parallel ultrasound, video, and audio data, of which approximately 13.5 hours are speech. This paper describes the corpus and presents benchmark results for the tasks of speech recognition, speech synthesis (articulatory-to-acoustic mapping), and automatic synchronisation of ultrasound to audio. The TaL corpus is publicly available under the CC BY-NC 4.0 license.

Ultrasound tongue imaging is used to visualise the intra-oral articulators during speech production. It is utilised in a range of applications, including speech and language therapy and phonetics research. Ultrasound and speech audio are recorded simultaneously, and in order to correctly use this data, the two modalities should be correctly synchronised. Synchronisation is achieved using specialised hardware at recording time, but this approach can fail in practice resulting in data of limited usability. In this paper, we address the problem of automatically synchronising ultrasound and audio after data collection. We first investigate the tolerance of expert ultrasound users to synchronisation errors in order to find the thresholds for error detection. We use these thresholds to define accuracy scoring boundaries for evaluating our system. We then describe our approach for automatic synchronisation, which is driven by a self-supervised neural network, exploiting the correlation between the two signals to synchronise them. We train our model on data from multiple domains with different speaker characteristics, different equipment, and different recording environments, and achieve an accuracy >92.4% on held-out in-domain data. Finally, we introduce a novel resource, the Cleft dataset, which we gathered with a new clinical subgroup and for which hardware synchronisation proved unreliable. We apply our model to this out-of-domain data, and evaluate its performance subjectively with expert users. Results show that users prefer our model’s output over the original hardware output 79.3% of the time. Our results demonstrate the strength of our approach and its ability to generalise to data from new domains.

Ultrasound tongue imaging (UTI) provides a convenient way to visualize the vocal tract during speech production. UTI is increasingly being used for speech therapy, making it important to develop automatic methods to assist various time-consuming manual tasks currently performed by speech therapists. A key challenge is to generalize the automatic processing of ultrasound tongue images to previously unseen speakers. In this work, we investigate the classification of phonetic segments (tongue shapes) from raw ultrasound recordings under several training scenarios: speaker-dependent, multi-speaker, speaker-independent, and speaker-adapted. We observe that models underperform when applied to data from speakers not seen at training time. However, when provided with minimal additional speaker information, such as the mean ultrasound frame, the models generalize better to unseen speakers.

Audiovisual synchronisation is the task of determining the time offset between speech audio and a video recording of the articulators. In child speech therapy, audio and ultrasound videos of the tongue are captured using instruments which rely on hardware to synchronise the two modalities at recording time. Hardware synchronisation can fail in practice, and no mechanism exists to synchronise the signals post hoc. To address this problem, we employ a two-stream neural network which exploits the correlation between the two modalities to find the offset. We train our model on recordings from 69 speakers, and show that it correctly synchronises 82.9% of test utterances from unseen therapy sessions and unseen speakers, thus considerably reducing the number of utterances to be manually synchronised. An analysis of model performance on the test utterances shows that directed phone articulations are more difficult to automatically synchronise compared to utterances containing natural variation in speech such as words, sentences, or conversations.

We investigate the automatic processing of child speech therapy sessions using ultrasound visual biofeedback, with a specific focus on complementing acoustic features with ultrasound images of the tongue for the tasks of speaker diarization and time-alignment of target words. For speaker diarization, we propose an ultrasound-based time-domain signal which we call estimated tongue activity. For word-alignment, we augment an acoustic model with low-dimensional representations of ultrasound images of the tongue, learned by a convolutional neural network. We conduct our experiments using the Ultrasuite repository of ultrasound and speech recordings for child speech therapy sessions. For both tasks, we observe that systems augmented with ultrasound data outperform corresponding systems using only the audio signal.

Paper number: 1736

This paper presents a simple count-based approach to learning word vector representations by leveraging statistics of cooccurrences between text and speech. This type of representation requires two discrete sequences of units defined across modalities. Two possible methods for the discretization of an acoustic signal are presented, which are then applied to fundamental frequency and energy contours of a transcribed corpus of speech, yielding a sequence of textual objects (e.g. words, syllables) aligned with a sequence of discrete acoustic events. Constructing a matrix recording the co-occurrence of textual objects with acoustic events and reducing its dimensionality with matrix decomposition results in a set of context-independent representations of word types. These are applied to the task of acoustic modelling for speech synthesis; objective and subjective results indicate that these representations are useful for the generation of acoustic parameters in a text-to-speech (TTS) system. In general, we observe that the more discretization approaches, acoustic signals, and levels of linguistic analysis are incorporated into a TTS system via these count-based representations, the better that TTS system performs.

We investigate two wavelet-based decomposition strategies of the f0 signal and their usefulness as a secondary task for speech synthesis using multi-task deep neural networks (MTL-DNN). The first decomposition strategy uses a static set of scales for all utterances in the training data. We propose a second strategy, where the scale of the mother wavelet is dynamically adjusted to the rate of each utterance. This approach is able to capture f0 variations related to the syllable, word, clitic-group, and phrase units. This method also constrains the wavelet components to be within the frequency range that previous experiments have shown to be more natural. These two strategies are evaluated as a secondary task in multi-task deep neural networks (MTL-DNNs). Results indicate that on an expressive dataset there is a strong preference for the systems using multi-task learning when compared to the baseline system.

We describe here a collection of speech data of bilingual and trilingual speakers of English, French, German and Italian. In the context of speech to speech translation (S2ST), this database is designed for several purposes and studies: training CLSA systems (cross-language speaker adaptation), conveying emphasis through S2ST systems, and evaluating TTS systems. More precisely, 36 speakers judged as accentless (22 bilingual and 14 trilingual speakers) were recorded for a set of 171 prompts in two or three languages, amounting to a total of 24 hours of speech. These sets of prompts include 100 sentences from news, 25 sentences from Europarl, the same 25 sentences with one acted emphasised word, 20 semantically unpredictable sentences, and finally a 240-word long text. All in all, it yielded 64 bilingual session pairs of the six possible combinations of the four languages. The database is freely available for non-commercial use and scientific research purposes.

A top-down hierarchical system based on deep neural networks is investigated for the modeling of prosody in speech synthesis. Suprasegmental features are processed separately from segmental features and a compact distributed representation of high-level units is learned at syllable-level. The suprasegmental representation is then integrated into a frame-level network. Objective measures show that balancing segmental and suprasegmental features can be useful for the frame-level network. Additional features incorporated into the hierarchical system are then tested. At the syllable-level, a bag-of-phones representation is proposed and, at the word-level, embeddings learned from text sources are used. It is shown that the hierarchical system is able to leverage new features at higher-levels more efficiently than a system which exploits them directly at the frame-level. A perceptual evaluation of the proposed systems is conducted and followed by a discussion of the results.

We propose a representation of f0 using the Continuous Wavelet Transform (CWT) and the Discrete Cosine Transform (DCT). The CWT decomposes the signal into various scales of selected frequencies, while the DCT compactly represents complex contours as a weighted sum of cosine functions. The proposed approach has the advantage of combining signal decomposition and higher-level representations, thus modeling low-frequencies at higher levels and high-frequencies at lower-levels. Objective results indicate that this representation improves f0 prediction over traditional short-term approaches. Subjective results show that improvements are seen over the typical MSD-HMM and are comparable to the recently proposed CWT-HMM, while using less parameters. These results are discussed and future lines of research are proposed.

The Continuous Wavelet Transform (CWT) has been recently proposed to model f0 in the context of speech synthesis.It was shown that systems using signal decomposition with the CWT tend to outperform systems that model the signal directly.The f0 signal is typically decomposed into various scales of differing frequency. In these experiments, we reconstruct f0 with selected frequencies and ask native listeners to judge the naturalness of synthesized utterances with respect to natural speech. Results indicate that HMM-generated f0 is comparable to the CWT low frequencies, suggesting it mostly generates utterances with neutral intonation. Middle frequencies achieve very high levels of naturalness, while very high frequencies are mostly noise.