Towards glottal source controllability in expressive speech synthesis

In order to obtain more human like sounding humanmachine interfaces we must first be able to give them expressive capabilities in the way of emotional and stylistic features so as to closely adequate them to the intended task. If we want to replicate those features it is not enough to merely replicate the prosodic information of fundamental frequency and speaking rhythm. The proposed additional layer is the modification of the glottal model, for which we make use of the GlottHMM parameters. This paper analyzes the viability of such an approach by verifying that the expressive nuances are captured by the aforementioned features, obtaining 95% recognition rates on styled speaking and 82% on emotional speech. Then we evaluate the effect of speaker bias and recording environment on the source modeling in order to quantify possible problems when analyzing multi-speaker databases. Finally we propose a speaking styles separation for Spanish based on prosodic features and check its perceptual significance.

Evaluation of a transplantation algorithm for expressive speech synthesis

When designing human-machine interfaces it is important to consider not only the bare bones functionality but also the ease of use and accessibility it provides. When talking about voice-based inter- faces, it has been proven that imbuing expressiveness into the synthetic voices increases signi?cantly its perceived naturalness, which in the end is very helpful when building user friendly interfaces. This paper proposes an adaptation based expressiveness transplantation system capable of copying the emotions of a source speaker into any desired target speaker with just a few minutes of read speech and without requiring the record- ing of additional expressive data. This system was evaluated through a perceptual test for 3 speakers showing up to an average of 52% emotion recognition rates relative to the natural voice recognition rates, while at the same time keeping good scores in similarity and naturality.

Towards speaking style transplantation in speech synthesis

One of the biggest challenges in speech synthesis is the production of naturally sounding synthetic voices. This means that the resulting voice must be not only of high enough quality but also that it must be able to capture the natural expressiveness imbued in human speech. This paper focus on solving the expressiveness problem by proposing a set of different techniques that could be used for extrapolating the expressiveness of proven high quality speaking style models into neutral speakers in HMM-based synthesis. As an additional advantage, the proposed techniques are based on adaptation approaches, which means that they can be used with little training data (around 15 minutes of training data are used in each style for this paper). For the final implementation, a set of 4 speaking styles were considered: news broadcasts, live sports commentary, interviews and parliamentary speech. Finally, the implementation of the 5 techniques were tested through a perceptual evaluation that proves that the deviations between neutral and speaking style average models can be learned and used to imbue expressiveness into target neutral speakers as intended.

New experiments on speaker diarization for unsupervised speaking style voice building for speech synthesis

El uso universal de síntesis de voz en diferentes aplicaciones requeriría un desarrollo sencillo de las nuevas voces con poca intervención manual. Teniendo en cuenta la cantidad de datos multimedia disponibles en Internet y los medios de comunicación, un objetivo interesante es el desarrollo de herramientas y métodos para construir automáticamente las voces de estilo de varios de ellos. En un trabajo anterior se esbozó una metodología para la construcción de este tipo de herramientas, y se presentaron experimentos preliminares con una base de datos multiestilo. En este artículo investigamos más a fondo esta tarea y proponemos varias mejoras basadas en la selección del número apropiado de hablantes iniciales, el uso o no de filtros de reducción de ruido, el uso de la F0 y el uso de un algoritmo de detección de música. Hemos demostrado que el mejor sistema usando un algoritmo de detección de música disminuye el error de precisión 22,36% relativo para el conjunto de desarrollo y 39,64% relativo para el montaje de ensayo en comparación con el sistema base, sin degradar el factor de mérito. La precisión media para el conjunto de prueba es 90.62% desde 76.18% para los reportajes de 99,93% para los informes meteorológicos.

A comparison of open-source segmentation architectures for dealing with imperfect data from the media in speech synthesis

Traditional Text-To-Speech (TTS) systems have been developed using especially-designed non-expressive scripted recordings. In order to develop a new generation of expressive TTS systems in the Simple4All project, real recordings from the media should be used for training new voices with a whole new range of speaking styles. However, for processing this more spontaneous material, the new systems must be able to deal with imperfect data (multi-speaker recordings, background and foreground music and noise), filtering out low-quality audio segments and creating mono-speaker clusters. In this paper we compare several architectures for combining speaker diarization and music and noise detection which improve the precision and overall quality of the segmentation.

Emotion transplantation through adaptation in HMM-based speech synthesis

This paper proposes an emotion transplantation method capable of modifying a synthetic speech model through the use of CSMAPLR adaptation in order to incorporate emotional information learned from a different speaker model while maintaining the identity of the original speaker as much as possible. The proposed method relies on learning both emotional and speaker identity information by means of their adaptation function from an average voice model, and combining them into a single cascade transform capable of imbuing the desired emotion into the target speaker. This method is then applied to the task of transplanting four emotions (anger, happiness, sadness and surprise) into 3 male speakers and 3 female speakers and evaluated in a number of perceptual tests. The results of the evaluations show how the perceived naturalness for emotional text significantly favors the use of the proposed transplanted emotional speech synthesis when compared to traditional neutral speech synthesis, evidenced by a big increase in the perceived emotional strength of the synthesized utterances at a slight cost in speech quality. A final evaluation with a robotic laboratory assistant application shows how by using emotional speech we can significantly increase the students’ satisfaction with the dialog system, proving how the proposed emotion transplantation system provides benefits in real applications.

Multilingual number transcription for text-to-speech conversion

This paper describes the text normalization module of a text to speech fully-trainable conversion system and its application to number transcription. The main target is to generate a language independent text normalization module, based on data instead of on expert rules. This paper proposes a general architecture based on statistical machine translation techniques. This proposal is composed of three main modules: a tokenizer for splitting the text input into a token graph, a phrase-based translation module for token translation, and a post-processing module for removing some tokens. This architecture has been evaluated for number transcription in several languages: English, Spanish and Romanian. Number transcription is an important aspect in the text normalization problem.

Development of a genre-dependent TTS system with cross-speaker speaking-style transplantation

One of the biggest challenges in speech synthesis is the production of contextually-appropriate naturally sounding synthetic voices. This means that a Text-To-Speech system must be able to analyze a text beyond the sentence limits in order to select, or even modulate, the speaking style according to a broader context. Our current architecture is based on a two-step approach: text genre identification and speaking style synthesis according to the detected discourse genre. For the final implementation, a set of four genres and their corresponding speaking styles were considered: broadcast news, live sport commentaries, interviews and political speeches. In the final TTS evaluation, the four speaking styles were transplanted to the neutral voices of other speakers not included in the training database. When the transplanted styles were compared to the neutral voices, transplantation was significantly preferred and the similarity to the target speaker was as high as 78%.

Synthesis of Nature Sounds for Speech Masking

The introduction of open-plan offices in the 1960s with the intent of making the workplace more flexible, efficient, and team-oriented resulted in a higher noise floor level, which not only made concentrated work more difficult, but also caused physiological problems, such as increased stress, in addition to a loss of speech privacy. Irrelevant background human speech, in particular, has proven to be a major factor in disrupting concentration and lowering performance. Therefore, reducing the intelligibility of speech and has been a goal of increasing importance in recent years. One method employed to do so is the use of masking noises, which consists in emitting a continuous noise signal over a loudspeaker system that conceals the perturbing speech. Studies have shown that while effective, the maskers employed to date – normally filtered pink noise – are generally poorly accepted by users. The collaborative "Private Workspace" project, within the scope of which this thesis was carried out, attempts to develop a coupled, adaptive noise masking system along with a physical structure to be used for open-plan offices so as to combat these issues. There is evidence to suggest that nature sounds might be more accepted as masker, in part because they can have a visual object that acts as the source for the sound. Direct audio recordings are not recommended for various reasons, and thus the nature sounds must be synthesized. This work done consists of the synthesis of a sound texture to be used as a masker as well as its evaluation. The sound texture is composed of two parts: a wind-like noise synthesized with subtractive synthesis, and a leaf-like noise synthesized through granular synthesis. Different combinations of these two noises produced five variations of the masker, which were evaluated at different levels along with white noise and pink noise using a modified version of an Oldenburger Satztest to test for an affect on speech intelligibility and a questionnaire to asses its subjective acceptance. The goal was to find which of the synthesized noises works best as a speech masker. This thesis first uses a theoretical introduction to establish the basics of sound perception, psychoacoustic masking, and sound texture synthesis. The design of each of the noises, as well as their respective implementations in MATLAB, is explained, followed by the procedures used to evaluate the maskers. The results obtained in the evaluation are analyzed. Lastly, conclusions are drawn and future work is and modifications to the masker are proposed. RESUMEN. La introducción de las oficinas abiertas en los años 60 tenía como objeto flexibilizar el ambiente laboral, hacerlo más eficiente y que estuviera más orientado al trabajo en equipo. Como consecuencia, subió el nivel de ruido de fondo, que no sólo dificulta la concentración, sino que causa problemas fisiológicos, como el aumento del estrés, además de reducir la privacidad. Hay estudios que prueban que las conversaciones de fondo en particular tienen un efecto negativo en el nivel de concentración y disminuyen el rendimiento de los trabajadores. Por lo tanto, reducir la inteligibilidad del habla es uno de los principales objetivos en la actualidad. Un método empleado para hacerlo ha sido el uso de ruido enmascarante, que consiste en reproducir señales continuas de ruido a través de un sistema de altavoces que enmascare el habla. Aunque diversos estudios demuestran que es un método eficaz, los ruidos utilizados hasta la fecha (normalmente ruido rosa filtrado), no son muy bien aceptados por los usuarios. El proyecto colaborativo "Private Workspace", dentro del cual se engloba el trabajo realizado en este Proyecto Fin de Grado, tiene por objeto desarrollar un sistema de ruido enmascarador acoplado y adaptativo, además de una estructura física, para su uso en oficinas abiertas con el fin de combatir los problemas descritos anteriormente. Existen indicios de que los sonidos naturales son mejor aceptados, en parte porque pueden tener una estructura física que simule ser la fuente de los mismos. La utilización de grabaciones directas de estos sonidos no está recomendada por varios motivos, y por lo tanto los sonidos naturales deben ser sintetizados. El presente trabajo consiste en la síntesis de una textura de sonido (en inglés sound texture) para ser usada como ruido enmascarador, además de su evaluación. La textura está compuesta de dos partes: un sonido de viento sintetizado mediante síntesis sustractiva y un sonido de hojas sintetizado mediante síntesis granular. Diferentes combinaciones de estos dos sonidos producen cinco variaciones de ruido enmascarador. Estos cinco ruidos han sido evaluados a diferentes niveles, junto con ruido blanco y ruido rosa, mediante una versión modificada de un Oldenburger Satztest para comprobar cómo afectan a la inteligibilidad del habla, y mediante un cuestionario para una evaluación subjetiva de su aceptación. El objetivo era encontrar qué ruido de los que se han sintetizado funciona mejor como enmascarador del habla. El proyecto consiste en una introducción teórica que establece las bases de la percepción del sonido, el enmascaramiento psicoacústico, y la síntesis de texturas de sonido. Se explica a continuación el diseño de cada uno de los ruidos, así como su implementación en MATLAB. Posteriormente se detallan los procedimientos empleados para evaluarlos. Los resultados obtenidos se analizan y se extraen conclusiones. Por último, se propone un posible trabajo futuro y mejoras al ruido sintetizado.