Can We Generate Emotional Pronunciations for Expressive Speech Synthesis?

Abstract

Abstract-In the field of expressive speech synthesis, a lot of work has been conducted on suprasegmental prosodic features while few has been done on pronunciation variants. However, prosody is highly related to the sequence of phonemes to be expressed. This article raises two issues in the generation of emotional pronunciations for TTS systems. The first issue consists in designing an automatic pronunciation generation method from text, while the second issue addresses the very existence of emotional pronunciations through experiments conducted on emotional speech. To do so, an innovative pronunciation adaptation method which automatically adapts canonical phonemes first to those labeled in the corpus used to create a synthetic voice, then to those labeled in an expressive corpus, is presented. This method consists in training conditional random fields pronunciation models with prosodic, linguistic, phonological and articulatory features. The analysis of emotional pronunciations reveals strong dependencies between prosody and phoneme assimilation or elisions. According to perceptual tests, the double adaptation allows to synthesize expressive speech samples of good quality, but emotion-specific pronunciations are too subtle to be perceived by testers.

Emotion Intensity and its Control for Emotional Voice Conversion

“…Emotion expressive speech is even more complex, which has subtle dynamic variations associated with multiple prosodic attributes [74], [75], [76]. Inspired by the successful attempts in prosody style control, several studies control the emotion intensity for emotional speech synthesis.…”

Réduction des segments en français spontané :apports des grands corpus et du traitement automatique de la parole

“…Il faut cependant garder à l'esprit que l'alignement forcé ne peut produire que des étiquetages en phones qui sont prévus par le dictionnaire de prononciation. L'alignement forcé et le dictionnaire de prononciation peuvent être utilisés pour étudier différentes hypothèses linguistiques et pour analyser de grands corpus (Adda-Decker et al, 1999 ;Boula de Mareüil et Adda-Decker, 2002 ;Van Bael et al, 2007 ;Schuppler et al, 2014 ;Wu et al, 2017 ;Tahon et al, 2018). Avec cette méthode, l'absence ou la présence du segment en question est décidée automatiquement par l'alignement forcé.…”

Expressiveness Influences Human Vocal Alignment Toward voice-AI

“…Recent developments in TTS synthesis have improved the acoustic parameters that influence degree of perceived expressiveness of the voices, cf. [3]- [5]. Thus, one question is whether humans will display alignment to TTS productions that are realized with increased acoustic-phonetic expressiveness, since it conveys robust human-like dynamism in the voice.…”

“…Some efforts to improve the perceived dynamism of TTS have focused on synthesizing acoustic expressiveness, based on human expressive vocal patterns, cf. [3]- [5]. Yet, how human users perceive and respond to these displays of expressiveness is an empirical question: do users respond to vocal expressiveness in voice-AI speech differently than non-expressive productions?…”