This paper describes a set of phone classification experiments based on electromyography (EMG) signals and a subsequent phone confusion analysis, as part of a project that aims to restore speech for Spanish laryngectomees by developing a Silent Speech Interface (SSI). Understanding the relationship between speech and the muscles used for speaking is essential to learn the possibilities and limitations of such EMG-based SSIs, before advancing to a complex task such as direct EMGto- speech conversion. When considering only information from the muscles of the face and neck, important information from the tongue and vocal cords is missing. This is reflected in the results, which show confusion between pairs of phones that only differ in the position of the tongue or the voicing feature.

A laryngectomee is a person whose larynx has been removed by surgery, usually due to laryngeal cancer. After surgery, most laryngectomees are able to speak again, using techniques that are learned with the help of a speech therapist. This is termed as alaryngeal speech, and esophageal speech (ES) is one of the several alaryngeal speech production modes. A considerable amount of research has been dedicated to the study of alaryngeal speech, with a wide range of aims such as helping speech therapists with evaluation and diagnosis, and improving its quality and intelligibility using digital signal processing techniques. We present to you a database of Spanish ES voices, named AhoSLABI, which is designed to allow the development of new support technologies for this speech impairment. The database primarily consists of recordings of 31 laryngectomees (27 males and 4 females) pronouncing phonetically balanced sentences. Additionally, it includes parallel recordings of the sentences by 9 healthy speakers (6 males and 3 females) to facilitate speech processing tasks that require small parallel corpora, such as voice conversion or synthetic speech adaptation. Apart from the sentences, the database includes sustained vowels and a small set of isolated words, which can be valuable for research on ES analysis, diagnosis and evaluation. The paper describes the main contents of the database, the recording protocols and procedure, as well as the labeling process. The main acoustic characteristics of the voices, such as speaking rate, durations of the recordings, phones and silences, and other such characteristics are compared with those of a reduced set of healthy voices. In addition, we describe an experiment using the database to improve the performance of an ASR system for ES speakers. This new resource will be made available to the scientific community with the hope that it will be used to improve the quality of life of the laryngectomees.

State of the art systems for voice conversion have been shown to generate highly natural sounding converted speech. Voice conversion techniques have also been applied to alaryngeal speech, with the aim of improving its quality or its intelligibility. In this paper, we present an attempt to apply a voice conversion strategy based on phonetic posteriorgrams (PPGs), which produces very high quality converted speech, to improve the characteristics of esophageal speech. The main advantage of this PPG based architecture lies in the fact that it is able to convert speech from any source, without the need to previously train the system with a parallel corpus. However, our results show that the PPG approach degrades the intelligibility of the converted speech considerably, especially when the input speech is already poorly intelligible. In this paper two systems are compared, an LSTM based one-to-one conversion system, which is referred to as the baseline, and the new system using phonetic posteriorgrams. Both spectral parameters and f0 are converted using DNN (Deep Neural Network) based architectures. Results from both objective and subjective evaluations are presented, showing that although ASR (Automated Speech Recognition) errors are reduced, original esophageal speech is still preferred by subjects.

This paper introduces the Synthetic Speech Detection system developed by Aholab for the Automatic Speaker Verification Spoofing and Countermeasures Challenge (ASVspoof 2015). The detector is a classifier based on Gaussian Mixture Models that are created using the Relative Phase Shift (RPS) transformation for the phase information. Different strategies have been evaluated: modeling the specific attacks using the information provided by the ASVspoof 2015 organizers, and modeling the vocoders possibly used in the spoofing signals, using data from previous works. The evaluation results show that attack specific models work for known attacks but they do not cope with the unknown attacks correctly. When using vocoder models build with other databases, the results suggest that the followed strategy do not take advantage of the available data and thus model adaptation should be explored. Index Terms: synthetic speech detection, phase information, anti-spoofing

In the field of speaker verification (SV) it is nowadays feasible and relatively easy to create a synthetic voice to deceive a speech driven biometric access system. This paper presents a synthetic speech detector that can be connected at the front-end or at the back-end of a standard SV system, and that will protect it from spoofing attacks coming from state-of-the-art statistical Text to Speech (TTS) systems. The system described is a Gaussian Mixture Model (GMM) based binary classifier that uses natural and copy-synthesized signals obtained from the Wall Street Journal database to train the system models. Three different state-of-the-art vocoders are chosen and modeled using two sets of acoustic parameters: 1) relative phase shift and 2) canonical Mel Frequency Cepstral Coefficients (MFCC) parameters, as baseline. The vocoder dependency of the system and multivocoder modeling features are thoroughly studied. Additional phase-aware vocoders are also tested. Several experiments are carried out, showing that the phase-based parameters perform better and are able to cope with new unknown attacks. The final evaluations, testing synthetic TTS signals obtained from the Blizzard challenge, validate our proposal.

Voice conversion methods based on frequency warping followed by amplitude scaling have been recently proposed. These methods modify the frequency axis of the source spectrum in such manner that some significant parts of it, usually the formants, are moved towards their image in the target speaker's spectrum. Amplitude scaling is then applied to compensate for the differences between warped source spectra and target spectra. This article presents a fully parametric formulation of a frequency warping plus amplitude scaling method in which bilinear frequency warping functions are used. Introducing this constraint allows for the conversion error to be described in the cepstral domain and to minimize it with respect to the parameters of the transformation through an iterative algorithm, even when multiple overlapping conversion classes are considered. The paper explores the advantages and limitations of this approach when applied to a cepstral representation of speech. We show that it achieves significant improvements in quality with respect to traditional methods based on Gaussian mixture models, with no loss in average conversion accuracy. Despite its relative simplicity, it achieves similar performance scores to state-of-the-art statistical methods involving dynamic features and global variance.

The importance of phase information in the perceptual quality of the speech signals is studied in this paper. Many speech synthesisers do not use the original phase information of the signals assuming their contribution is almost inaudible. The Relative Phase Shift (RPS) representation of the phase allows straightforward phase structure analysis, manipulation and resynthesis, and we use these features to do a comparative evaluation of some phase modifications usually found in speech models. The final intention of this study is to get an answer to the question of whether phases deserve elaborate models to get high quality synthetic speech, or their subtle effects justify overlooking them. Index Terms: phase perception, RPS, speech synthesis. 1.

In this paper, we evaluate the vulnerability of speaker verification (SV) systems to synthetic speech. The SV systems are based on either the Gaussian mixture model–universal background model (GMM-UBM) or support vector machine (SVM) using GMM supervectors. We use a hidden Markov model (HMM)-based text-to-speech (TTS) synthesizer, which can synthesize speech for a target speaker using small amounts of training data through model adaptation of an average voice or background model. Although the SV systems have a very low equal error rate (EER), when tested with synthetic speech generated from speaker models derived from the Wall Street Journal (WSJ) speech corpus, over 81% of the matched claims are accepted. This result suggests vulnerability in SV systems and thus a need to accurately detect synthetic speech. We propose a new feature based on relative phase shift (RPS), demonstrate reliable detection of synthetic speech, and show how this classifier can be used to improve security of SV systems.

Currently, the statistical framework based on Hidden Markov Models (HMMs) plays a relevant role in speech synthesis, while voice conversion systems based on Gaussian Mixture Models (GMMs) are almost standard. In both cases, statistical modeling is applied to learn distributions of acoustic vectors extracted from speech signals, each vector containing a suitable parametric representation of one speech frame. The overall performance of the systems is often limited by the accuracy of the underlying speech parameterization and reconstruction method. The method presented in this paper allows accurate MFCC extraction and high-quality reconstruction of speech signals assuming a Harmonics plus Noise Model (HNM). Its suitability for high-quality HMM-based speech synthesis is shown through subjective tests.

In this paper, we present new results from our research into the vulnerability of a speaker verification (SV) system to synthetic speech. We use a HMM-based speech synthesizer, which creates synthetic speech for a targeted speaker through adaptation of a background model and both GMM-UBM and support vector machine (SVM) SV systems. Using 283 speakers from the Wall-Street Journal (WSJ) corpus, our SV systems have a 0.35% EER. When the systems are tested with synthetic speech generated from speaker models derived from the WSJ journal corpus, over 91% of the matched claims are accepted. We propose the use of relative phase shift (RPS) in order to detect synthetic speech and develop a GMM-based synthetic speech classifier (SSC). Using the SSC, we are able to correctly classify human speech in 95% of tests and synthetic speech in 88% of tests thus significantly reducing the vulnerability.

Statistical parametric synthesizers have achieved very good performance scores during the last years. Nevertheless, as they require the use of vocoders to parameterize speech (during training) and to reconstruct waveforms (during synthesis), the speech generated from statistical models lacks some degree of naturalness. In previous works we explored the usefulness of the harmonics plus noise model in the design of a high-quality speech vocoder. Quite promising results were achieved when this vocoder was integrated into a synthesizer. In this paper, we describe some recent improvements related to the excitation parameters, particularly the so called maximum voiced frequency. Its estimation and explicit modelling leads to an even better synthesis performance as confirmed by subjective comparisons with other well-known methods.

Voice conversion has been traditionally focused on spectrum. Current systems lack a solid prosody conversion method suitable for different speaking styles. Recently, the unit selection technique has been applied to transform emotional intonation contours. This paper goes one step beyond: it explores strategies for training and configuring the selection cost function in an emotion conversion application. The proposed system, which uses accent groups as basic intonation units and performs conversion also on phoneme durations and intensity, is evaluated by means of a carefully designed subjective test involving the big six emotions. Although the expressiveness of the converted sentences is still far from that of natural emotional speech, satisfactory results are obtained when different configurations are used for different emotions.

The definition of parameters is a crucial step in the development of a system for identifying emotions in speech. Although there is no agreement on which are the best features for this task, it is generally accepted that prosody carries most of the emotional information. Most works in the field use some kind of prosodic features, often in combination with spectral and voice quality parametrizations. Nevertheless, no systematic study has been done comparing these features. This paper presents the analysis of the characteristics of features derived from prosody, spectral envelope, and voice quality as well as their capability to discriminate emotions. In addition, early fusion and late fusion techniques for combining different information sources are evaluated. The results of this analysis are validated with experimental automatic emotion identification tests. Results suggest that spectral envelope features outperform the prosodic ones. Even when different parametrizations are combined, the late fusion of long-term spectral statistics with short-term spectral envelope parameters provides an accuracy comparable to that obtained when all parametrizations are combined.

A novel algorithm based on classical cepstrum calculation followed by dynamic programming is presented in this paper. The algorithm has been evaluated with a 60-minutes database containing 60 speakers and different recording conditions and environments. A second reference database has also been used. In addition, the performance of four popular PDA algorithms has been evaluated with the same databases. The results prove the good performance of the described algorithm in noisy conditions. Furthermore, the paper is a first initiative to perform an evaluation of widely used PDA algorithms over an extensive and realistic database.

Building a text corpus suitable to be used in corpus-based speech synthesis is a time-consuming process that usually requires some human intervention to select the desired phonetic content and the necessary variety of prosodic contexts. If an emotional text-to-speech (TTS) system is desired, the complexity of the corpus generation process increases. This paper presents a study aiming to validate or reject the use of a semantically neutral text corpus for the recording of both neutral and emotional (acted) speech. The use of this kind of texts would eliminate the need to include semantically emotional texts into the corpus. The study has been performed for Basque language. It has been made by performing subjective and objective comparisons between the prosodic characteristics of recorded emotional speech using both semantically neutral and emotional texts. At the same time, the performed experiments allow for an evaluation of the capability of prosody to carry emotional information in Basque language. Prosody manipulation is the most common processing tool used in concatenative TTS. Experiments of automatic recognition of the emotions considered in this paper (the "Big Six emotions") show that prosody is an important emotional indicator, but cannot be the only manipulated parameter in an emotional TTS system-at least not for all the emotions. Resynthesis experiments transferring prosody from emotional to neutral speech have also been performed. They corroborate the results and support the use of a neutral-semantic-content text in databases for emotional speech synthesis.

This paper presents the experiments made to automatically identify emotion in an emotional speech database for Basque. Three different classifiers have been built: one using spectral features and GMM, other with prosodic features and SVM and the last one with prosodic features and GMM. 86 prosodic features were calculated and then an algorithm to select the most relevant ones was applied. The first classifier gives the best result with a 98.4 % accuracy when using 512 mixtures, but the classifier built with the best 6 prosodic features achieves an accuracy of 92.3 % in spite of its simplicity, showing that prosodic information is very useful to identify emotions. 1.

Abstract 1 In this work we present a telephone speech database for Basque, compliant with the guidelines of the Speechdat project. The database contains 1060 calls from the fixed telephone network. We first describe the main aspects of the database design. We also present the recognition results using the database and a set of procedures following the language independent reference recogniser commonly named Refrec. 1.

The present paper presents the modeling of standard Basque intonation to be used in text to speech conversion systems. The parameterization process of the Basque f0 curves made according to Fujisaki's intonation model is explained: experiments made in the placing of the phrase commands of the model are described and the results of these experiments are analyzed. The statistical analysis of the obtained parameters using classification and regression trees is described and the results obtained in this study are also explained.

International audience