Speech Recognition In Noisy Environments Research Articles

Strategies of hearing aids fitting for improving speech recognition in noisy environments

To assess speech recognition in users of mono- and binaural hearing aids (HA) with dynamic environment control system (DECS). 25 HA users with symmetrical bilateral moderate to severe sensorineural hearing loss and 10 individuals with normal hearing underwent speech audiometry in a free sound field by listening to an adaptive Russian matrix sentence test with a background noise. The testing was conducted for hearing impaired patients with one and two HA. The Speech Recognition Thresholds in noise (SRTN) using HA with DESC and using HA without DESC were compared. The dichotic digits test and the rapidly alternating speech perception test were performed to evaluate the central auditory system function. The SRTN in normal hearing listeners was -16.4±1.9 dB SNR. In the case of monaural using HA without DESC the SRTN was -2.5±4.4 dB SNR, using HA with DESC it was -5.8±4.0 dB SNR; in the case of using binaural HAs it was -6.3±3.8 dB SNR and -9.9±3.1 dB SNR accordingly. HAs with Dynamic Environment Control System are highly effective for speech perception in noisy environments. Binaural HA use is more effective than monaural HA use regardless of HA model. Speech recognition in noisy environments in cases of binaural HA use correlates significantly with results of tests assessing the central auditory system function.

DNN-Based Mask Estimation for Distributed Speech Enhancement in Spatially Unconstrained Microphone Arrays

Deep neural network (DNN)-based speech enhancement algorithms in microphone arrays have now proven to be efficient solutions to speech understanding and speech recognition in noisy environments. However, in the context of ad-hoc microphone arrays, many challenges remain and raise the need for distributed processing. In this paper, we propose to extend a previously introduced distributed DNN-based time-frequency mask estimation scheme that can efficiently use spatial information in form of so-called compressed signals which are pre-filtered target estimations. We study the performance of this algorithm named Tango under realistic acoustic conditions and investigate practical aspects of its optimal application. We show that the nodes in the microphone array cooperate by taking profit of their spatial coverage in the room. We also propose to use the compressed signals not only to convey the target estimation but also the noise estimation in order to exploit the acoustic diversity recorded throughout the microphone array.

Hindi speech recognition in noisy environment using hybrid technique

Automatic speech recognition is generally analyzed for two types of word utterances; isolated and continuous-words speech. Continuous-words speech is almost natural way of speaking but is difficult to be recognized through machines (speech recognizers). It is also highly sensitive to environmental variations. There are various parameters which are directly affecting the performance of automatic speech recognition like size of datasets/corpus, type of data sets (isolated, spontaneous or continuous) and environment variations (noisy/clean). The performance of speech recognizers is generally good in clean environments for isolated words, but it becomes typical in noisy environments especially for continuous words/sentences and is still a challenge. In this paper, a hybrid feature extraction technique is proposed by joining core blocks of PLP (perceptual linear predictive) and Mel frequency cepstral coefficients (MFCC) that can be utilized to improve the performance of speech recognizers under such circumstances. Voice activity and detection (VAD)-based frame dropping formula has been used solely within the training part of ASR (automatic speech recognition) procedure obviating its need in actual implementations. The motivation to use this formula is for removal of pauses and distorted elements of speech improving the phonemes modeling further. The proposed method shows average improvement in performance by 12.88% for standard datasets.

Effect of Forward Masking on Frequency Following Response as a Function of Age.

Forward masking occurs when noise is presented before the target signal, making the latter difficult to be perceived. It is related to temporal auditory processing and consequently to speech recognition in noisy environments, which may decline with age. Interest in forward masking has grown in the last years. Studies investigate psychoacoustic and electrophysiological recordings in different age- groups. The purpose of the study was to investigate the effect of forward masking on frequency following response (FFR) as a function of age. Cross-sectional analytical observational study. We assessed 69 normal-hearing participants of both genders assigned to three groups: 40 young individuals (aged 18-25 years, mean age = 22 years 8 months), 21 middle-age individuals (aged 25-55 years, mean age = 37 years 2 months), and 8 seniors (aged <55 years, mean age = 65 years 3 months). FFRs were recorded using the /da/ syllable with and without noise. The /da/ syllable and speech-shaped noise were monaurally presented to the participants' right ears through ER-3a insert earphones. Electrodes were placed in M1 and M2 (-), Fz (+), and Fpz (ground). Acquisition occurred under two conditions: (1) the/da/ syllable presented without the noise and (2) the /da/ syllable presented 4 msec after the noise. Data show that (1) considering the mean values of all participants, there was a significant latency delay of all waves (PV, A, PW, PX, PY, PZ, and O) when the /da/ syllable was presented 4 msec after the masking noise as compared with the condition without noise, that is, forward masking occurred in all components of the FFR responses, and (2) for the youngest group and the middle-age group, forward masking was seen for all waves, except PX in the latter one; for the senior group, an irregular pattern was observed (presence of forward masking in PA, PY, PZ, and O). This pattern may be due to an aging effect on FFR responses even without noise presence, which makes it more difficult to identify forward masking effect in this population. Although it is well documented in the literature that forward masking increases with age, this is less evident on FFR recordings in the senior population. An aging effect was identified in FFR responses. Forward masking was identified in FFR responses of all groups but less evident in senior population.

An ensemble technique for speech recognition in noisy environments

&lt;span&gt;Automatic speech recognition (ASR) is a technology that allows a computer and mobile device to recognize and translate spoken language into text. ASR systems often produce poor accuracy for the noisy speech signal. Therefore, this research proposed an ensemble technique that does not rely on a single filter for perfect noise reduction but incorporates information from multiple noise reduction filters to improve the final ASR accuracy. The main factor of this technique is the generation of K-copies of the speech signal using three noise reduction filters. The speech features of these copies differ slightly in order to extract different texts from them when processed by the ASR system. Thus, the best among these texts can be elected as final ASR output. The ensemble technique was compared with three related current noise reduction techniques in terms of CER and WER. The test results were encouraging and showed a relatively decreased by 16.61% and 11.54% on CER and WER compared with the best current technique. ASR field will benefit from the contribution of this research to increase the recognition accuracy of a human speech in the presence of background noise.&lt;/span&gt;

The effect of a coding strategy that removes temporally masked pulses on speech perception by cochlear implant users

Speech recognition in noisy environments remains a challenge for cochlear implant (CI) recipients. Unwanted charge interactions between current pulses, both within and between electrode channels, are likely to impair performance. Here we investigate the effect of reducing the number of current pulses on speech perception. This was achieved by implementing a psychoacoustic temporal-masking model where current pulses in each channel were passed through a temporal integrator to identify and remove pulses that were less likely to be perceived by the recipient. The decision criterion of the temporal integrator was varied to control the percentage of pulses removed in each condition. In experiment 1, speech in quiet was processed with a standard Continuous Interleaved Sampling (CIS) strategy and with 25, 50 and 75% of pulses removed. In experiment 2, performance was measured for speech in noise with the CIS reference and with 50 and 75% of pulses removed. Speech intelligibility in quiet revealed no significant difference between reference and test conditions. For speech in noise, results showed a significant improvement of 2.4 dB when removing 50% of pulses and performance was not significantly different between the reference and when 75% of pulses were removed. Further, by reducing the overall amount of current pulses by 25, 50, and 75% but accounting for the increase in charge necessary to compensate for the decrease in loudness, estimated average power savings of 21.15, 40.95, and 63.45%, respectively, could be possible for this set of listeners. In conclusion, removing temporally masked pulses may improve speech perception in noise and result in substantial power savings.

Convolutional neural network acoustic model for robust Indonesian speech recognition in noisy environment

Noise causes the decreasing accuracy of automatic speech recognition (ASR). Several techniques have been developed and proposed to overcome this problem. Using artificial neural network (ANN) as acoustic model is one of the techniques. Convolutional neural network (CNN) is a variant of ANN that has been used for acoustic modeling. Another approach is to do pre-processing to the speech signal or to the extracted acoustic feature from speech signal, such as cepstral mean and variance normalization (CMVN). On this work, CNN acoustic models were trained by using CMVN pre-processed acoustic feature to make a noise-robust speech recognition system. Two group of models were made, each to handle 2 kinds of noise (babble noise and street noise). Those acoustic models were tested with noisy speech at different SNR (signal-to-noise ratio) value. Testing results from CNN acoustic models were compared with the ones from Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) acoustic models. Testing results showed the increasing accuracy scores of acoustic models when models were trained using more variation of training data. CNN acoustic models that were trained using FBANK feature have higher accuracy scores than GMM-HMM models that were built using the same feature.

변형된 고조파 복원 방법을 이용한 잡음환경에서의 음성인식

변형된 고조파 복원 방법을 이용한 잡음환경에서의 음성인식

Cochlear implantation as a treatment for single-sided deafness and asymmetric hearing loss: a randomized controlled evaluation of cost-utility

BackgroundSingle-sided deafness (SSD) and asymmetric hearing loss (AHL) have recently been proposed as a new indication for cochlear implantation. There is still no recommended treatment for these hearing deficits, and most options considered rely on the transfer of sound from the poor ear to the better ear, using Contralateral Routing of the Signal (CROS) hearing aids or bone conduction (BC) devices. In contrast, cochlear implantation allows the poor ear to be stimulated and binaural hearing abilities to be partially restored. Indeed, most recently published studies have reported an improvement in the spatial localisation of an incoming sound and better speech recognition in noisy environments after cochlear implantation in SSD/AHL subjects. It also provides consistent relief of tinnitus when associated. These encouraging hearing outcomes raise the question of the cost-utility of this expensive treatment in an extended indication.MethodsThe final endpoint of this national multicentre study is to determine the incremental cost-utility ratio (ICUR) of cochlear implantation in comparison to the current standard of care in France through simple observation, using a randomised controlled trial. Firstly, the study comprises a prospective and descriptive part, where 150 SSD/AHL subjects try CROS hearing aids and a BC device for three weeks each. Secondly, the choice is made between CROS hearing aids, BC implanted device and cochlear implantation. Hearing outcomes and quality of life measurements are described after 6 months for the subjects who chose CROS, BC or declined any option. The subjects who opt for cochlear implantation are randomised between one group where the cochlear implant is inserted without delay and one group of simple initial observation. Hearing outcomes and quality of life measurements are compared after 6 months.DiscussionThe present study was designed to assess the efficiency of cochlear implantation in SSD/AHL. A favourable cost-utility ratio in this extended indication would strengthen the promising clinical results and justify a reimbursement by the health insurance. The efficiency of other options (CROS, BC) will also be described.Trial registrationThis research has been registered in ClinicalTrials.gov (http://www.clinicaltrials.gov/), the 29th July 2014 under the n°NCT02204618.

Speech enhancement method based on feature compensation gain for effective speech recognition in noisy environments

Speech enhancement method based on feature compensation gain for effective speech recognition in noisy environments

Effect of Forward Masking on Frequency Following Response as a Function of Age

Background: Forward masking occurs when noise is presented before the target signal, making thelatter difficult to be perceived. It is related to temporal auditory processing and consequently to speechrecognition in noisy environments, which may decline with age. Interest in forward masking has grown inthe last years. Studies investigate psychoacoustic and electrophysiological recordings in different age groups.&lt;br /&gt;Purpose: The purpose of the study was to investigate the effect of forward masking on frequency followingresponse (FFR) as a function of age.&lt;br /&gt;Research Design: Cross-sectional analytical observational study.&lt;br /&gt;Study Sample: We assessed 69 normal-hearing participants of both genders assigned to three groups: 40young individuals (aged 18–25 years,mean age = 22 years, 8 months), 21 middle-age individuals (aged 25–55years, mean age = 37 years, 2 months), and 8 seniors (aged &lt; 55 years, mean age = 65 years, 3 months).&lt;br /&gt;Intervention: FFRs were recorded using the /da/ syllable with and without noise.&lt;br /&gt;Data Collection and Analysis: The /da/ syllable and speech-shaped noise were monaurally presentedto the participants’ right ears through ER-3a insert earphones. Electrodes were placed in M1 and M2 (-),Fz (+), and Fpz (ground). Acquisition occurred under two conditions: (1) the /da/ syllable presented withoutthe noise and (2) the /da/ syllable presented 4 msec after the noise.&lt;br /&gt;Results: Data show that (1) considering the mean values of all participants, there was a significant latencydelay of all waves (PV, A, PW, PX, PY, PZ, and O) when the /da/ syllable was presented 4 msecafter the masking noise as compared with the condition without noise, that is, forward masking occurredin all components of the FFR responses, and (2) for the youngest group and the middle-age group, forwardmasking was seen for all waves, except PX in the latter one; for the senior group, an irregular patternwas observed (presence of forward masking in PA, PY, PZ, and O). This pattern may be due to an agingeffect on FFR responses even without noise presence, which makes it more difficult to identify forwardmasking effect in this population. Although it is well documented in the literature that forward maskingincreases with age, this is less evident on FFR recordings in the senior population.&lt;br /&gt;Conclusions: An aging effect was identified in FFR responses. Forward masking was identified in FFRresponses of all groups but less evident in senior population.&lt;br /&gt;

Rank-1 constrained Multichannel Wiener Filter for speech recognition in noisy environments

Multichannel linear filters, such as the Multichannel Wiener Filter (MWF) and the Generalized Eigenvalue (GEV) beamformer are popular signal processing techniques which can improve speech recognition performance. In this paper, we present an experimental study on these linear filters in a specific speech recognition task, namely the CHiME-4 challenge, which features real recordings in multiple noisy environments. Specifically, the rank-1 MWF is employed for noise reduction and a new constant residual noise power constraint is derived which enhances the recognition performance. To fulfill the underlying rank-1 assumption, the speech covariance matrix is reconstructed based on eigenvectors or generalized eigenvectors. Then the rank-1 constrained MWF is evaluated with alternative multichannel linear filters under the same framework, which involves a Bidirectional Long Short-Term Memory (BLSTM) network for mask estimation. The proposed filter outperforms alternative ones, leading to a 40% relative Word Error Rate (WER) reduction compared with the baseline Weighted Delay and Sum (WDAS) beamformer on the real test set, and a 15% relative WER reduction compared with the GEV-BAN method. The results also suggest that the speech recognition accuracy correlates more with the Mel-frequency cepstral coefficients (MFCC) feature variance than with the noise reduction or the speech distortion level.

The use of the Frequency Modulation System by hearing-impaired children: benefits from the family's perspective.

To evaluate the family's perspective of benefits of the frequency modulation (FM) system adapted to children with sensorineural hearing loss. This is a descriptive, analytical, cross-sectional study with the participation of family members of hearing-impaired children aged 6 to 15 years, users of hearing aids and/or cochlear implants, benefited with the FM system by a hearing health service of the Unified Health System (SUS), Brazil. The FM Listening Evaluation For Children questionnaire with 14 questions was used to evaluate the benefits of using the FM system, the characteristics of the hearing aids and/or cochlear implants, and the brand/model of the FM system the children used. Statistically significant differences were found between the questionnaire responses of FM users and non-users in all situational analyses (quiet, noise, auditory only, distance), with better learning performance and improvements and attention among FM users, from the families' point of view. No method was used to formally evaluate such performances. It was observed that, according to the parents' and guardians' perceptions, the use of FM systems improves the performance of hearing-impaired children in various acoustic situations, with special emphasis on speech recognition in noisy environments and at increased distances from the sound source. Although used in different contexts and situations, the FM system has brought greater benefits for the children in the school environment, according to the respondents.

Research progress on cochlear implantation in patients with unilateral deafness

Cochlear implantation (CI) is a relatively new therapeutic method for patients with severe-to-profound unilateral sensorineural hearing loss. Acoustic signal can stimulate the ear directly, thus patients can gain better binaural hearing. This passage has reviewed the recent researches about CI for patients suffered from single side deafness. It can improve the ability of sound localization and speech recognition in noisy environment for patients, in addition, it may relieve tinnitus to some degree. But the long-term effect and indication selection still need further study.

An experimental framework for Arabic digits speech recognition in noisy environments

In this paper we present an experimental framework for Arabic isolated digits speech recognition named ARADIGITS-2. This framework provides a performance evaluation of Modern Standard Arabic devoted to a Distributed Speech Recognition system, under noisy environments at various Signal-to-Noise Ratio (SNR) levels. The data preparation and the evaluation scripts are designed by deploying a similar methodology to that followed in AURORA-2 database. The original speech data contains a total of 2704 clean utterances, spoken by 112 (56 male and 56 female) Algerian native speakers, down-sampled at 8 kHz. The feature vectors, which consist of a set of Mel Frequency Cepstral Coefficients and log energy, are extracted from speech samples using ETSI Advanced Front-End (ETSI-AFE) standard; whereas, the Hidden Markov Models (HMMs) Toolkit is used for building the speech recognition engine. The recognition task is conducted in speaker-independent mode by considering both word and syllable as acoustic units. Therefore, an optimal fitting of HMM parameters, as well as the temporal derivatives window, is carried out through a series of experiments performed on the two training modes: clean and multi-condition. Better results are obtained by exploiting the polysyllabic nature of Arabic digits. These results show the effectiveness of syllable-like unit in building Arabic digits recognition system, which exceeds word-like unit by an overall Word Accuracy Rate of 0.44 and 0.58% for clean and multi-condition training modes, respectively.

Speech perception and quality of life of open-fit hearing aid users.

ABSTRACTObjective To relate the performance of individuals with hearing loss at high frequencies in speech perception with the quality of life before and after the fitting of an open-fit hearing aid (HA).Methods The WHOQOL-BREF had been used before the fitting and 90 days after the use of HA. The Hearing in Noise Test (HINT) had been conducted in two phases: (1) at the time of fitting without an HA (situation A) and with an HA (situation B); (2) with an HA 90 days after fitting (situation C).Study Sample Thirty subjects with sensorineural hearing loss at high frequencies.Results By using an analysis of variance and the Tukey’s test comparing the three HINT situations in quiet and noisy environments, an improvement has been observed after the HA fitting. The results of the WHOQOL-BREF have showed an improvement in the quality of life after the HA fitting (paired t-test). The relationship between speech perception and quality of life before the HA fitting indicated a significant relationship between speech recognition in noisy environments and in the domain of social relations after the HA fitting (Pearson’s correlation coefficient).Conclusions The auditory stimulation has improved speech perception and the quality of life of individuals.

Feature Extraction Method for Improving Speech Recognition in Noisy Environments

The paper presents a feature extraction method, named as Normalized Gammachirp Cepstral Coefficients (NGCC) that incorporates the properties of the peripheral auditory system to improve robustness in noisy speech recognition. The proposed method is based on a second order low-pass filter and normalized gammachirp filterbank to emulate the mechanisms performed in the outer/middle ear and cochlea. The speech recognition performance of this method is conducted on the speech signals in real-world noisy environments. Experimental results demonstrate that method outperformed the classical feature extraction methods in terms of speech recognition rate. The used Hidden Markov Models based speech recognition system is employed on the HTK 3.4.1 platform (Hidden Markov Model Toolkit).

Roles of the Contralateral Efferent Reflex in Hearing Demonstrated with Cochlear Implants.

Our two ears do not function as fixed and independent sound receptors; their functioning is coupled and dynamically adjusted via the contralateral medial olivocochlear efferent reflex (MOCR). The MOCR possibly facilitates speech recognition in noisy environments. Such a role, however, is yet to be demonstrated because selective deactivation of the reflex during natural acoustic listening has not been possible for human subjects up until now. Here, we propose that this and other roles of the MOCR may be elucidated using the unique stimulus controls provided by cochlear implants (CIs). Pairs of sound processors were constructed to mimic or not mimic the effects of the contralateral MOCR with CIs. For the non-mimicking condition (STD strategy), the two processors in a pair functioned independently of each other. When configured to mimic the effects of the MOCR (MOC strategy), however, the two processors communicated with each other and the amount of compression in a given frequency channel of each processor in the pair decreased with increases in the output energy from the contralateral processor. The analysis of output signals from the STD and MOC strategies suggests that in natural binaural listening, the MOCR possibly causes a small reduction of audibility but enhances frequency-specific inter-aural level differences and the segregation of spatially non-overlapping sound sources. The proposed MOC strategy could improve the performance of CI and hearing-aid users.

Speech Intelligibility in Various Noise Conditions with the Nucleus® 5 CP810 Sound Processor.

The Nucleus® 5 System Sound Processor (CP810, Cochlear™, Macquarie University, NSW, Australia) contains two omnidirectional microphones. They can be configured as a fixed directional microphone combination (called Zoom) or as an adaptive beamformer (called Beam), which adjusts the directivity continuously to maximally reduce the interfering noise. Initial evaluation studies with the CP810 had compared performance and usability of the new processor in comparison with the Freedom™ Sound Processor (Cochlear™) for speech in quiet and noise for a subset of the processing options. This study compares the two processing options suggested to be used in noisy environments, Zoom and Beam, for various sound field conditions using a standardized speech in noise matrix test (Oldenburg sentences test). Nine German-speaking subjects who previously had been using the Freedom speech processor and subsequently were upgraded to the CP810 device participated in this series of additional evaluation tests. The speech reception threshold (SRT for 50% speech intelligibility in noise) was determined using sentences presented via loudspeaker at 65 dB SPL in front of the listener and noise presented either via the same loudspeaker (S0N0) or at 90 degrees at either the ear with the sound processor (S0NCI+) or the opposite unaided ear (S0NCI-). The fourth noise condition consisted of three uncorrelated noise sources placed at 90, 180 and 270 degrees. The noise level was adjusted through an adaptive procedure to yield a signal to noise ratio where 50% of the words in the sentences were correctly understood.In spatially separated speech and noise conditions both Zoom and Beam could improve the SRT significantly. For single noise sources, either ipsilateral or contralateral to the cochlear implant sound processor, average improvements with Beam of 12.9 and 7.9 dB in SRT were found. The average SRT of –8 dB for Beam in the diffuse noise condition (uncorrelated noise from both sides and back) is truly remarkable and comparable to the performance of normal hearing listeners in the same test environment. The static directivity (Zoom) option in the diffuse noise condition still provides a significant benefit of 5.9 dB in comparison with the standard omnidirectional microphone setting. These results indicate that CI recipients may improve their speech recognition in noisy environments significantly using these directional microphone-processing options.

Optimum integration weight for decision fusion audio-visual speech recognition

Automatic speech recognition (ASR) technologies have been successfully applied to several real world applications. But, still there exist several problems that need to be solved for wider application of the technologies. One problem is noise-robustness of recognition performance. Recently, audio-visual speech recognition (AVSR) has received attention as a solution to this problem. In this, visual speech information is used together with acoustic signal for speech recognition in noisy environments. This paper presents a new decision fusion AVSR system, in which the classifier's decision is optimised using genetic algorithm (GA) optimisation technique. Hence, the optimally fused decision fusion AVSR system produces robust recognition accuracy at all SNR conditions. For evaluating the performance of the proposed scheme, the recognition results are compared with those of an equal weight bimodal AVSR system and with another state-of-the-art method, namely, compression and Mel sub-band spectral subtraction (CMSBS)-based noise compensation method for speech recognition in noise. Further, to show the effectiveness of the proposed optimisation method, the recognition results are compared with those of a similar method called directed grid search method, which also optimises the integration weight against the recognition accuracy.