Vahid Montazeri

Two-channel multi-stage speech enhancement for noisy fMRI environment

Strong acoustic noise interferes with speech communication during functional magnetic resonance imaging (fMRI) scans employing echo planar imaging (EPI) sequences. Performance of many speech enhancement (SE) methods degrades with such a low signal-to-noise ratio (SNR). In this paper, an adaptive SE method is proposed enabling noise-free online speech communication with patients and recording of clean speech. The proposed method is a two-channel, two-stage algorithm with a switching technique. In the first stage, least mean square (LMS) based algorithm and linear prediction error filtering, for specific parts of the noisy acoustic signal, are used. The first stage operations improve the performance of minimum mean square error log-spectral amplitude estimator (Log-MMSE) used for enhancing the speech signal in the second stage. A soft-switching technique is also employed which significantly reduces the audible noise spikes which occur during the transient periods between successive EPI sequencing. The proposed algorithm is tested online during the scans in a noisy fMRI room. Moreover, this paper provides the results of real time implementation of the algorithm on a dedicated laboratory test-bed setup using Texas Instruments digital signal processing platform.

Evaluation of a new approach for speech enhancement algorithms in hearing aids

Several studies on hearing impaired people who use hearing aid reveal that speech enhancement algorithms implemented in hearing aids improve listening comfort. However, these algorithms do not improve speech intelligibility too much and in many cases they decrease the speech intelligibility, both in hearing-impaired and in normally hearing people. In fact, current approaches for development of the speech enhancement algorithms (e.g. minimum mean square error (MMSE)) are not optimal for intelligibility improvement. Some recent studies investigated the effect of different distortions on the enhanced speech and realized that by controlling the amplification distortion, the intelligibility improves dramatically. In this paper, we examined, subjectively and objectively, the effects of amplification distortion on the speech enhanced by two algorithms in three background noises at different SNR levels.

Single channel speech enhancement technique for low SNR quasi-periodic noise based on reduced order linear prediction

In this paper we propose an efficient single microphone (single channel) speech enhancement (SE) method for Quasi-Stationary noise environment. The proposed method estimates the noise by exploiting its quasi-periodic nature, followed by a statistical model based method to enhance the speech. An efficient reduced-order linear predictive error filtering is introduced to increase the signal to noise ratio (SNR) of the noisy speech. The proposed method is evaluated experimentally by considering the actual recorded Functional Magnetic Resonance Imaging (fMRI) machinery noise which is quasi periodic in nature, added in clean speech. Objective evaluation of our method shows improvement in both quality and intelligibility measures when tested with the sentences chosen from IEEE corpus added in broadband quasi periodic fMRI noise. The proposed method outperforms the standard statistical model based SE technique.

Precedence based speech segregation in bilateral cochlear implant users

The precedence effect (PE) enables the perceptual dominance by a source (lead) over an echo (lag) in reverberant environments. In addition to facilitating sound localization, the PE can play an important role in spatial unmasking of speech. Listeners attending to binaural vocoder simulations with identical channel center frequencies and phase demonstrated PE-based benefits in a closed-set speech segregation task. When presented with the same stimuli, bilateral cochlear implant users did not derive such benefits. These findings suggest that envelope extraction in itself may not lead to a breakdown of the PE benefits, and that other factors may play a role.

Perceiving foreign-accented speech with decreased spectral resolution in single- and multiple-talker conditions

To determine the effect of reduced spectral resolution on the intelligibility of foreign-accented speech, vocoder-processed sentences from native and Mandarin-accented English talkers were presented to listeners in single- and multiple-talker conditions. Reduced spectral resolution had little effect on native speech but lowered performance for foreign-accented speech, with a further decrease in multiple-talker conditions. Following the initial exposure, foreign-accented speech with reduced spectral resolution was less intelligible than unprocessed speech in both single- and multiple-talker conditions. Intelligibility improved with extended exposure, but only for single-talker conditions. Results indicate a perceptual impairment when perceiving foreign-accented speech with reduced spectral resolution.

Constraints on ideal binary masking for the perception of spectrally-reduced speech

This study investigated recognition of sentences processed using ideal binary masking (IBM) with limited spectral resolution. Local thresholds (LCs) of -12, 0, and 5 dB were applied which altered the target and masker power following IBM. Recognition was reduced due to persistence of the masker and limited target recovery, thus preventing IBM from ideal target-masker segregation. Linear regression and principal component analyses showed that, regardless of masker type and number of spectral channels, higher LCs were associated with poorer recognition. In addition, limitations on target recovery led to more detrimental effects on speech recognition compared to persistence of the masker.

Predicting speech reception thresholds of cochlear implant users using a modified envelope based measure

Investigation of a model to predict cochlear implant users intelligibility in noise.Participants aided hearing loss and model predictions were weakly correlated.A modified model which adjusts the model predictions based on aided hearing loss.Higher prediction accuracy using the modified model.The new model generalizes to listeners with a wider range of audiometric thresholds. An envelope-correlation based measure (ECM), proposed in Yousefian and Loizou (2012), was applied to predict the speech reception thresholds (SRTs) of nine pre- and post-lingual cochlear implant (CI) users in the presence of speech-shaped masker and a female competing talker. A principal component analysis showed that CI users aided pure-tone thresholds were almost orthogonal to the predictions made by ECM, indicating that the ECM does not take aided hearing loss into account. A modified ECM is proposed which includes pure-tone thresholds into the predictions. The proposed model adjusts the ECM predictions based on the pure tone threshold values, and is able to predict CI users SRTs with higher correlation (r=0.92) compared to the original ECM (r=0.59). The results reported in this study indicate that information available on MAPs (e.g. T-levels, number of active electrodes) may not be sufficient to predict CI users SRTs in noise. In additions, results show that lack of audibility (i.e. elevated pure-tone thresholds) can dramatically limit speech intelligibility in noise in (even experienced) CI users.

Effects of residual masker, spectral resolution, and low frequency cues on the perception of ideal binary-masked speech

This study investigated the intelligibility of ideal binary-masked (IdBM) stimuli under conditions with limited spectral resolution. In experiment 1, we used different IdBM local thresholds, which retain different amounts of residual masker and target information in the IdBM-stimuli. The stimuli were presented to 30 normal hearing listeners. The results indicated that, with a 6- or 12-channel tone-vocoder, the presence of residual masker in the IdBM-stimuli limits the intelligibility scores, thus preventing IdBM processing from achieving an ideal segregation of target from masker. In experiment 2, we investigated whether introduction of low frequency target information to the tone-vocoded IdBM-stimuli improves the intelligibility scores. Twenty normal hearing listeners participated in this experiment. The results indicated that in contrast to target F0 cues, inclusion of low-pass filtered target information (cutoff = 300 Hz) helped listeners segregate the target from the masker, thus improving the intelligibility of the IdBM-stimuli. These results argue against F0 as a segregation cue in electroacoustic conditions and suggest that target F0 cues do not provide benefits beyond those provided by IdBM.This study investigated the intelligibility of ideal binary-masked (IdBM) stimuli under conditions with limited spectral resolution. In experiment 1, we used different IdBM local thresholds, which retain different amounts of residual masker and target information in the IdBM-stimuli. The stimuli were presented to 30 normal hearing listeners. The results indicated that, with a 6- or 12-channel tone-vocoder, the presence of residual masker in the IdBM-stimuli limits the intelligibility scores, thus preventing IdBM processing from achieving an ideal segregation of target from masker. In experiment 2, we investigated whether introduction of low frequency target information to the tone-vocoded IdBM-stimuli improves the intelligibility scores. Twenty normal hearing listeners participated in this experiment. The results indicated that in contrast to target F0 cues, inclusion of low-pass filtered target information (cutoff = 300 Hz) helped listeners segregate the target from the masker, thus improving the intelli...

Effect of frequency shifts on talker recognition in native and foreign-accented speech

The present study examined the effect of frequency shifts on perceived talker recognition in foreign-accented speech compared to native-accented speech. Sentences were processed using the STRAIGHT vocoder. The spectral envelope and the fundamental frequency were shifted up or down in seven steps (3 up, 3 down plus unshifted) using scale factors of 8% and 30%, respectively, at each step. Listeners heard pairs of sentences and were asked to judge whether the identity of the talker was the same or different. Frequency shifts had similar effects for native- and foreign-accent conditions, in that listeners perceived the shifted versions as different talkers when, in fact, the talkers were the same. However, listeners were more likely to judge native-accented sentence pairs as the same talker regardless of whether or not they were the same; foreign-accented sentence pairs were more likely to be heard as different talkers. Overall, these results indicate that patterns of frequency-shifted foreign-accented speech...

Perception of voice gender in children's voices by cochlear implant users

Previous research [Assmann et al., J. Acoust Soc. Am 138, 1811 (2015)] investigated normal-hearing (NH) listeners’ ability to discriminate age and gender in children’s speech. Speech stimuli (/hVd/ syllables from 140 speakers between 5 and 18 years of age) were processed using STRAIGHT to simulate a change in perceived gender. Experimental conditions involved swapping the fundamental frequency (F0) contour and/or formant frequencies (FF) to the opposite-sex average at each age level. This research was extended by presenting the stimuli to cochlear implant (CI) users. Preliminary results from two CI users have led to two main conclusions. First, whereas NH listeners used both F0 and FF to discriminate voice gender, CI users relied primarily on F0. Second, NH listeners and CI users demonstrated differential patterns of voice gender misclassification, particularly in the case of young children. NH listeners, while frequently making errors, identified a majority of young boys as male and young girls as female...