Deep-learning-based audio-visual speech enhancement in presence of Lombard effect

Abstract

Keywords:Lombard effect audio-visual speech enhancement deep learning speech quality speech intelligibility A B S T R A C T When speaking in presence of background noise, humans reflexively change their way of speaking in order to improve the intelligibility of their speech. This reflex is known as Lombard effect. Collecting speech in Lombard conditions is usually hard and costly. For this reason, speech enhancement systems are generally trained and evaluated on speech recorded in quiet to which noise is artificially added. Since these systems are often used in situations where Lombard speech occurs, in this work we perform an analysis of the impact that Lombard effect has on audio, visual and audio-visual speech enhancement, focusing on deep-learning-based systems, since they represent the current state of the art in the field.We conduct several experiments using an audio-visual Lombard speech corpus consisting of utterances spoken by 54 different talkers. The results show that training deep-learning-based models with Lombard speech is beneficial in terms of both estimated speech quality and estimated speech intelligibility at low signal to noise ratios, where the visual modality can play an important role in acoustically challenging situations. We also find that a performance difference between genders exists due to the distinct Lombard speech exhibited by males and females, and we analyse it in relation with acoustic and visual features. Furthermore, listening tests conducted with audio-visual stimuli show that the speech quality of the signals processed with systems trained using Lombard speech is statistically significantly better than the one obtained using systems trained with non-Lombard speech at a signal to noise ratio of −5 dB. Regarding speech intelligibility, we find a general tendency of the benefit in training the systems with Lombard speech.

Evaluation of Lombard Speech Models in the Context of Speech in Noise Enhancement

“…As seen from this discussion, not only the experimental setup was different compared to our approach, but the analysis also differs from that performed by us; thus a direct comparison is not possible. Even the observation of what SNR value the model does not work at seems to be uncommon; in the case of our research, the models stop working at a threshold of -5 dB, in the work of Michelsanti et al [57], it refers to 5 dB.…”

“…The ESTOI values changed dramatically from 0.442 for -20 dB to -5 dB SNR, up to 0.927 for a SNR range between 10 and 30 dB. So, the relative performance of the systems at SNR ≤ 5 dB is similar to that observed for the systems trained on a narrow SNR range [57].…”

“…When comparing the obtained results to the state-of-theart, one can see that such a comparison in practice is not straightforward. For example, Michelsanti et al [57] reported averaged scores of PESQ and ESTOI (Extended Short-Time Objective Intelligibility) [58] measures for a deep-learningbased system of audio-visual speech enhancement with the Lombard effect applied. To elicit the Lombard effect, Speech Shaped Noise (SSN) at 80 dB Sound Pressure Level (SPL) was presented to the speakers while they were reading the sentences [57].…”

“…For example, Michelsanti et al [57] reported averaged scores of PESQ and ESTOI (Extended Short-Time Objective Intelligibility) [58] measures for a deep-learningbased system of audio-visual speech enhancement with the Lombard effect applied. To elicit the Lombard effect, Speech Shaped Noise (SSN) at 80 dB Sound Pressure Level (SPL) was presented to the speakers while they were reading the sentences [57]. It is worth noting that ESTOI scores, which estimate speech intelligibility, range from 0 to 1, where high values correspond to high speech intelligibility.…”

“…Results obtained by Michelsanti et al [57] refer to two types of subjective tests, namely the MUSHRA and speech intelligibility tests. For an AO-L at -5 dB SNR, the result was approx.…”