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

Abstract

This paper investigates the temporal excitation patterns of creaky voice. Creaky voice is a voice quality frequently used as a phrase-boundary marker, but also as a means of portraying attitude, affective states and even social status. Consequently, the automatic detection and modelling of creaky voice may have implications for speech technology applications. The acoustic characteristics of creaky voice are, however, rather distinct from modal phonation. Further, several acoustic patterns can bring about the perception of creaky voice, thereby complicating the strategies used for its automatic detection, analysis and modelling. The present study is carried out using a variety of languages, speakers, and on both read and conversational data and involves a mutual information-based assessment of the various acoustic features proposed in the literature for detecting creaky voice. These features are then exploited in classification experiments where we achieve an appreciable improvement in detection accuracy compared to the state of the art. Both experiments clearly highlight the presence of several creaky patterns. A subsequent qualitative and quantitative analysis of the identified patterns is provided, which reveals a considerable speaker-dependent variability in the usage of these creaky patterns. We also investigate how creaky voice detection systems perform across creaky patterns.

Tracheoesophageal speech: A dedicated objective acoustic assessment

“…It turns out that speakers in the TTS and Control sets have a creaky usage never exceeding 10%. This goes in line with the findings in (Drugman et al, 2013a) drawn over a range of languages (US English, Japanese, Swedish and Finnish) where creaky voice was used between 3.5 and 10.5% of the time (as extracted from manual annotations). The artefacts in TE speech however lead to a perception of creakiness largely exceeding 10% for about 3 patients over 4.…”

A Log Domain Pulse Model for Parametric Speech Synthesis

“…To model voiced and unvoiced time-frequency regions, most vocoders rely on an f 0 estimate and/or voicing detection that assume voiced segments to have sinusoidal content. However, various segments of the speech signal are voiced with very non-periodical characteristic of the pulse's position, called creakiness in this presentation, as in creaky voice phonatory mode [36] and sometimes in transients. Thus, the corresponding sinusoidal content in these segments is highly disturbed and often wrongly classified as unvoiced segments, leading to hoarseness and noisy transients in the synthesized voice.…”

Characterisation of voice quality of Parkinson’s disease using differential phonological posterior features

“…Recent work on non-modal phonation focuses on detection (Drugman et al, 2014), analysis (Malyska, 2008;Malyska et al, 2011) and synthesis (Bangayan et al, 1997) of speech with non-modal phonation. Modern computational paralinguistics tries to 1) get rid of non-modal phonation, or 2) model it, for example, for classification purposes (Schuller and Batliner, 2013).…”