Reference-free automatic quality assessment of tracheoesophageal speech

Abstract

Abstract-Evaluation of the quality of tracheoesophageal (TE) speech using machines instead of human experts can enhance the voice rehabilitation process for patients who have undergone total laryngectomy and voice restoration. Towards the goal of devising a reference-free TE speech quality estimation algorithm, we investigate the efficacy of speech signal features that are used in standard telephone-speech quality assessment algorithms, in conjunction with a recently introduced speech modulation spectrum measure. Tests performed on two TE speech databases demonstrate that the modulation spectral measure and a subset of features in the standard ITU-T P.563 algorithm estimate TE speech quality with better correlation (up to 0.9) than previously proposed features.

Tracheoesophageal speech: A dedicated objective acoustic assessment

“…In the context of the original feature definition, an additional noise was found to be caused by room reverberation (Falk et al, 2010). For TE speech, it was shown in (Huang et al, 2009) that these additional modulation frequencies are due to artefacts (e.g. gargling noise, raspiness) found in the voice.…”

“…gargling noise, raspiness) found in the voice. In (Huang et al, 2009), RSMR was observed to be the most efficient feature for the automatic estimation of the TE speech quality. Figure 6 displays, for the three datasets, the distributions of the proportion of creakiness (on a logarithmic scale) and of RSMR.…”

“…• Use of A4S to predict a non-automatic susbjective assessment: Similarly to what has been achieved in (Huang et al, 2009) or (McDonald et al, 2010, the acoustic features proposed in this manuscript could be used to predict subjective ratings of the TE voice quality. Furthermore, since A4S characterizes various aspects of the voice, it could be interesting to investigate the correlation of its axes with perceptual dimensions such as those used in the GRBAS (De Bodt et al, 1997) or in the IINFVo (Moerman et al, 2006) scales.…”

“…In addition to these, the work described in (Huang et al, 2009) investigated the use of existing acoustic analysis techniques, originally developed for the analysis of narrowband telephone speech, in order to estimate the TE speech quality. Results showed that a subset of the features from the International Telecommunication Union recommendation P.563 algorithm estimate the TE speech quality with better correlation than previously proposed features (Huang et al, 2009). With the same goal in mind, the Moore-Glasberg auditory model was employed in (McDonald et al, 2010) to extract perceptually relevant features from the acoustic signal.…”