Jędrzej Kociński

Speech intelligibility improvement using convolutive blind source separation assisted by denoising algorithms

The present study is concerned with the blind source separation (BSS) of speech and speech-shaped noise sources. All recordings were carried out in an anechoic chamber using a dummy head (two microphones, one in each ear). The program which implements the algorithm for BSS of convolutive mixtures introduced by Parra and Spence [Parra, L., Spence, C., 2000a. Convolutive blind source separation of non-stationary sources. IEEE Trans. Speech Audio Process. 8(3), 320-327 (US Patent US6167417)] was used to separate out the signals. In the postprocessing phase two different denoising algorithms were used. The first was based on a minimum mean-square error log-spectral amplitude estimator [Ephraim, E., Malah, D., 1985. Speech enhancement using a minimum mean-square error log-spectral amplitude estimator. IEEE Trans. Speech Audio Process. ASSP-33(2), 443-445], while the second one was based on Wiener filter in which the concept of an a priori signal-to-noise estimation presented by Ephraim (as mentioned above) was applied [Scalart, P., Filho, J.V., 1996. Speech enhancement based on a priori signal to noise estimation. IEEE Internat. Conf. Acoust. Speech Signal Process. 1, 629-632]. Non-sense word tests were used as a target speech in both cases while one or two disturbing sources were used as interferences. The speech intelligibility before and after the BSS was measured for three subjects with audiologically normal hearing. Next the speech signal after BSS was denoised and presented to the same listeners. The results revealed some ambiguities caused by the insufficient number of microphones compared to the number of sound sources. For one disturbance only, the intelligibility improvement was significant. However, when there were two disturbances and the target speech, the separation was much poorer. The additional denoising, as could be expected, raises the intelligibility slightly. Although the BSS method requires more research on optimization, the results of the investigation imply that it may be applied to hearing aids in the future.

Audio-visual interaction of environmental noise

Currently research into the psychological evaluation of noise in daily life is carried out without taking into account the sense of sight. The human senses interact with each other; thus some information coming from one sense can be skipped or ignored in favor of information coming from another sense, leading to completely different reactions or behavior. The aim of this paper is to verify, on the basis of psychophysical experiments, how a human being processes audio-visual information coming from the different environmental noises which can be encountered in daily life. The experiment was divided into three parts: auditory, visual, and audio-visual. In each part of the experiment, the ICBEN scale (0-10) was used to rate the presented stimuli. In the first part only audio stimuli were presented, and subjects were asked to rate their annoyance with the sound. In the second part of the experiment, the participants were asked to rate how pleasant the presented video clips were. Finally, in the last part of the experiment, participants were presented with a compatible and incompatible mix of audio and visual stimuli and asked to rate their annoyance. We found that several audio stimuli were assessed differently, to a significant extent, by listeners after video clips were added to them.

Spatial efficiency of blind source separation based on decorrelation - subjective and objective assessment

Blind source separation (BSS) method is one of the newest multisensorial methods that exploits statistical properties of simultaneously recorded independent signals to separate them out. The objective of this method is similar to that of beamforming, namely a set of spatial filters that separate source signals are calculated. Thus, it seems to be reasonable to investigate the spatial efficiency of BSS that is reported in this study. A dummy head with two microphones was used to record two signals in an anechoic chamber: target speech and babble noise in different spatial configurations. Then the speech reception thresholds (SRTs, i.e. signal-to-noise ratio, SNR yielding 50% speech intelligibility) before and after BSS algorithm (Parra and Spence, 2000) were determined for audiologically normal subjects. A significant speech intelligibility improvement was noticed after the BSS was applied. This happened in most cases when the target and masker sources were spatially separated. Moreover, the comparison of objective (SNR enhancement) and subjective (intelligibility improvement) assessment methods is reported here. It must be emphasized that these measures give different results.

The interaction between speech intelligibility task and non-auditory tasks

A very sensitive method of speech intelligibility measurement was used.The influence of non-auditory task on speech intelligibility was not noticed.The influence of speech task on non-auditory task performance was noticed.The results confirmed the cross-modal perceptual load effect. Many research on various aspects of speech perception and intelligibility has been conducted so far. However, the ability of human mind to perform multiple tasks simultaneously is still a challenging topic. Perceptual load theory states that people first decide what activity is most important after which available perceptual resources are allocated. Less important activities are conducted effectively only if remaining resources are sufficient. It should be emphasized that researchers studying divided attention focused mainly on activities requiring vision. This paper focuses on interaction between non-auditory tasks (distracters) and a speech intelligibility task. A very sensitive method based on speech reception threshold (SRT) measurement was used instead of standard speech intelligibility determination. This sensitivity is a result of a very steep psychometric function (ipso facto small standard deviation) which is not possible to achieve for standard methods. Therefore, this property can be used to investigate even slight impact of external conditions on speech perception. Two experiments examining the interactions between these two tasks were conducted. Despite the use of such sensitive method, in none of the experiments speech intelligibility decreased due to distracter. However, the percentage of correct responses to the distracter when subjects were performing speech intelligibility tests decreased compared to the situation when they focused only on the distracter.