An Auditory Neural Feature Extraction Method for Robust Speech Recognition

Combined speech enhancement and auditory modelling for robust distributed speech recognition

Recently, Li et al. proposed a new auditory feature for robust speech recognition in noise environments. The new feature was derived by mimicking closely the function of human auditory process. Several filters were used to model the outer ear, middle ear, and cochlea, and the initial filter parameters and shapes were obtained from crude psychoacoustics results, experience, or experiments. Although one may adjust the feature parameters by hand to get better performance, the resulting feature parameters still may not be optimal in the sense of minimal recognition errors, especially for different tasks. To further improve the auditory feature, in this paper we apply discriminative training to optimize the auditory feature parameters with some guidance from psychoacoustic evidence but otherwise in a data-driven approach so as to minimize the recognition errors. One significant contribution over similar efforts in the past, such as discriminative feature extraction, is that we make no assumption on the parametric form of the auditory filters. Instead, we only require the filters to be smooth and triangular-like as suggested by psychoacoustics research. Our approach is evaluated on the Aurora database and achieves a word error reduction of 19.2%.

Most current speech recognition systems are built upon a single type of model, e.g. an HMM or certain type of segment based model, and furthermore typically employs only one type of acoustic feature e.g. MFCCs and their variants. This entails that the system may not be robust should the modeling assumptions be violated. Recent research efforts have investigated the use of multi-scale/multi-band acoustic features for robust speech recognition. This paper described a multi-model approach as an alternative and complement to the multi-feature approaches. The multi-model approach seeks a combination of different types of acoustic models, thereby integrating the capabilities of each individual model for capturing discriminative information. An example system built upon the combination of the standard HMM technique with a segment-based modeling technique was implemented. Experiments for both isolated-word and continuous speech recognition have shown improved performances over each of the individual models considered in isolation.

This paper presents an investigation of robust features for speech recognition in three different noisy environments. The state-of art Mel-frequency cepstral coefficients were extensively explored in additive, convolutive and reverberant environments. These environments have captured the interest of many researches in speech recognition systems. We evaluate robust speech recognition results on the TI-DIGIT database. Significant word error rate reductions were observed in the connected digit recognition experiments. The recognition experiments vindicate the robustness of Mel-frequency cepstral coefficient with dynamic features and cepstral mean normalization in hostile environments, especially additive and reverberant noise

Intra-speaker variations due to perceptually induced stress or emotion adversely affect speech recognition system performance. In this paper, we combine auditory based (Mel frequency cepstral coefficients and linear predictive cepstral coefficients) features and articulatory based (voicedness) features for robust speech recognition. Voicedness features are derived using linear and teager energy operator (TEO) based nonlinear fast Fourier transform (FFT) spectra. Nonlinear properties are analyzed in both the time and frequency domains. In addition, we investigate the sensitivity of all these FFT spectra to stress and observe the performance of individual FFT spectra. The system is tested using stressed speech data from the speech under simulated and actual stress (SUSAS) database. The results show that articulatory based features help to improve the system performance. Furthermore, significant performance improvement has been observed when using the FFT spectrum which is less sensitive to stress

S.5492-5495

Exploring joint equalization of spatial-temporal contextual statistics of speech features for robust speech recognition

In this paper, we propose two estimators for the autocorrelation sequence of a periodic signal in additive noise. Both estimators are formulated employing tables which contain all the possible products of sample pairs in a speech signal frame. The first estimator is based on a pitch-synchronous averaging. This estimator is statistically analyzed and we show that the signal-to-noise ratio (SNR) can be increased up to a factor equal to the number of available periods. The second estimator is similar to the former one but it avoids the use of those sample products more likely affected by noise. We prove that, under certain conditions, this estimator can remove the effect of an additive noise in a statistical sense. Both estimators are employed to extract mel frequency cepstral coefficients (MFCCs) as features for robust speech recognition. Although these estimators are initially conceived for voiced speech frames, we extend their application to unvoiced sounds in order to obtain a coherent feature extractor. The experimental results show the superiority of the proposed approach over other MFCC-based front-ends such as the higher-lag autocorrelation spectrum estimation (HASE), which also employs the idea of avoiding those autocorrelation coefficients more likely affected by noise.

It is well known that the performance of a speech recognizer, trained with clean speech database, usually degrades drastically when operating in noisy environments. To make it robust to noise, a novel model adaptation method based on the SNR-dependent non-linear spectral compression MFCC features is proposed, which modifies the HMMs of the clean speech to generate the adaptation models on the basis of the estimated Mel-band SNR and the spectral compression coefficients. In this approach, only the clean speech models are adopted, which not only adapt and compensate the means but also modify the variances, hence it can deal with much lower SNR. In addition, this approach can be used in variable noisy environments. Such methods for adaptive approach will be discussed in detail in this paper. By the adoption of the new adaptive approach, substantial improvement can be observed in recognizing in different noisy environments.

<p style="text-align: justify;"><span lang="EN-US">Objectives</span><span lang="EN-US">. Children with Autism Spectrum Disorder (ASD) often face difficulties in speech perception, particularly in noisy environments. These problems are associated with central auditory processing disorders. FM systems, which improve the signal-to-noise ratio, can be used to enhance speech perception. The aim of this study was to investigate the effects of FM systems when working with children with ASD in noisy conditions and their use in a school setting. Methods. The pilot study was conducted in two stages. In the first stage, the “Words in Noise” test and the “Sentence Repetition” task were used to assess children’s ability to recognize speech in noisy conditions with and without the use of FM systems. Fourteen children with ASD and 14 typically developing peers participated in the experiment. In the second stage, FM systems were studied in real school environments with 10 elementary school students with autism. Changes in auditory abilities were assessed using the L.I.F.E.-R scale. Results. In the first stage, it was found that children with ASD recognized significantly fewer words in noisy conditions compared to their typically developing peers. In the “Sentence Repetition” task, the use of FM systems improved the performance of children with ASD from 58.3% to 76.9% (p=0.0005). In the second stage, most participants showed minimal changes in auditory ability scores on the L.I.F.E.-R scale: the average score before using FM systems was 54.9, and after using them it was 57</span><span lang="EN-US">.4 (p=0.2322). However, several students showed improvements of 8–13 points, reflecting individual variability in the effect of using FM systems. Discussion. FM systems have demonstrated their effectiveness in improving speech perception in noisy environments for children with ASD. However, the integration of FM systems into the school process requires further research on their effectiveness in real conditions, as well as adaptations to minimize discomfort for children and improve interaction with teachers and tutors.</span></p>

Automatic Speech Recognition has been an active topic of research for the past four decades. The main objective of the automatic speech recognition task is to convert a speech segment into an interpretable text message without the need of human intervention. Many different algorithms and schemes based on different mathematical paradigms have been proposed in an attempt to improve recognition rates. Cepstral coefficients play an important part in speech theory and in automatic speech recognition in particular due to their ability to compactly represent relevant information that is contained in a short time sample of a continuous speech signal. The goal of this paper is to discuss comparison of speech parameterization methods: Mel-Frequency Cepstrum Coefficients (MFCC) and improved Mel-Frequency Cepstrum Coefficients (MFCC) using RASTA filters. Thus, in this study, we try to improve the MFCC algorithms to achieve much accuracy reducing the error rates in Automatic Speech Recognition. First, we remove signal correlation through normalization, then we use RASTA filter to filtering the cepstral coefficients. Finally, we reduce dimension of the cepstral coefficients by the variances of cepstral coefficients in different dimension and obtain our features. By using various classifiers, we try to simulate the speech feature extraction at much optimal and least error rate providing robust method for Automatic Speech Recognition (ASRs).KeywordsAutomatic Speech RecognitionMel frequency Cepstrum Coefficients (MFCC’s)ERB Gammatone FilteringHidden Markov Model

This paper presents a word graph based feature enhancement method for robust speech recognition in noise. The approach uses signal processing based speech enhancement as a starting point, and then performs Wiener filtering to remove residual noise. During the process, a decoded word graph is used to directly guide the feature enhancement with respect to the HMM for recognition, so that the enhanced feature can match the clean speech model better in the acoustic space. The proposed word graph based feature enhancement method was tested on the Aurora 2 database. Experimental results show that an improved recognition performance can be obtained comparing with conventional signal processing based and GMM based feature enhancement methods. With signal processing based weighted noise estimation and GMM based method, the relative error rate reductions are 35.44% and 42.58%, respectively. The proposed word graph based method improves the performance further, and a relative error rate reduction of 57.89% is obtained.

SNR loss: A new objective measure for predicting the intelligibility of noise-suppressed speech

Cochlear implant (CI) recipients generally have good perception of speech in quiet environments but difficulty perceiving speech in noisy conditions, reduced sensitivity to speech prosody, and difficulty appreciating music. Auditory training has been proposed as a method of improving speech perception for CI recipients, and recent efforts have focussed on the potential benefits of music-based training. This study evaluated two melodic contour training programs and their relative efficacy as measured on a number of speech perception tasks. These melodic contours were simple 5-note sequences formed into 9 contour patterns, such as “rising” or “rising-falling.” One training program controlled difficulty by manipulating interval sizes, the other by note durations. Sixteen adult CI recipients (aged 26–86 years) and twelve normal hearing (NH) adult listeners (aged 21–42 years) were tested on a speech perception battery at baseline and then after 6 weeks of melodic contour training. Results indicated that there were some benefits for speech perception tasks for CI recipients after melodic contour training. Specifically, consonant perception in quiet and question/statement prosody was improved. In comparison, NH listeners performed at ceiling for these tasks. There was no significant difference between the posttraining results for either training program, suggesting that both conferred benefits for training CI recipients to better perceive speech.

Cochlear implant (CI) users struggle to understand speech in noisy conditions. To address this problem, we propose a deep learning speech denoising sound coding strategy that estimates the CI electric stimulation patterns out of the raw audio data captured by the microphone, performing end-to-end CI processing. To estimate the relative denoising performance differences between various approaches, we compared this technique to a classic Wiener filter and to a convTasNet. Speech enhancement performance was assessed by means of signal-to-noise-ratio improvement and the short-time objective speech intelligibility measure. Additionally, 5 CI users were evaluated for speech intelligibility in noise to assess the potential benefits of each algorithm. Our results show that the proposed method is capable of replacing a CI sound coding strategy while preserving its general use for every listener and performing speech enhancement in noisy environments, without sacrificing algorithmic latency.