Prediction of creaky voice from contextual factors

Abstract

Creaky voice, also referred to as vocal fry, is a voice quality frequently produced in many languages, in both read and conversational speech. In order to enhance the naturalness of speech synthesisers, these latter should be able to generate speech in all its expressive diversity. This includes a proper use of creaky voice. The goal of this paper is two-fold. Firstly we analyse how contextual factors can be informative for the prediction of creaky use. It is observed that a few contextual factors related to speech production preceding a silence or a pause are of particular interest. This study validates that creaky voice plays a crucial syntactic role, allowing for a better structuring of phrases. In a second experiment, we investigate the prediction of creakiness from contextual factors based on HMMs. Four methods are compared on a US English and a Finnish speaker. It is shown that the best prediction technique achieves a promising performance comparable to what is carried out with the creaky detection algorithm on which HMMs were trained.

Data-driven detection and analysis of the patterns of creaky voice

“…Creaky voice has been studied in relation to various functions in speech communication, and most commonly with phrase or sentence boundary marking (Surana and Slifka, 2006b;Drugman et al, 2013). Similarly creaky voice has been associated with turn-yielding in Finnish (Ogden, 2001).…”

HMM-based synthesis of creaky voice

“…Synthesis of voice with creak requires i) the prediction of creaky parts from context and ii) the ability to render creaky excitation. In our previous work, we have developed methods for creaky voice prediction from context [13] and rendering of creaky excitation [12]. However, these methods have not been utilised in a full TTS voice before.…”

“…In this study, the algorithm in [5] is used, which provides a frame-wise probability of creak. This parameter is used as a feature in the HMMtraining for determining if a segment is creaky or not [13]. More specifically, the parameter indicating the probability of creak is trained as an additional 1-dimensional feature along with other speech features, such as F0 and spectrum.…”

“…For the Finnish speaker, MV, a total of 66 contextual factors are used, described in [26]. According to the study in [13], only a few of the contextual factors are useful in predicting creaky voice, and the useful factors are closely related with creaky use at the end of a sentence or a word group.…”

“…Further work by the present authors was concerned with developing an excitation model of creaky production capable of providing a natural rendering of the voice quality [12]. Also the prediction of creaky voice from contextual factors was investigated in [13], which enables automatic determination of the creaky usage from the input text. One obvious application of this line of research is incorporating creaky voice in a statistical parametric speech synthesis system.…”