Search

Postfiltering is a well known technique that helps increasing the quality of coded speech, enhancing speech intelligibility or alleviating the oversmoothing effect of statistical speech processing methods. This letter presents a new formulation of the radial cepstral postfiltering method that enables the application of different postfiltering factors to low and high frequencies. The transition between bands will be smooth and controllable through an adjustable cut-off frequency. The proposed algorithm can be implemented by means of a simple multiplicative matrix of which an analytical expression is derived. The new method provides a flexible framework to tackle issues related to the quality and intelligibility of synthetic speech.

An improved speech recognition algorithm has been developed with additional signal processing units to increase the number of comparative data parameters and to reduce the natural and spectral distortion of speech signals. Using a verified database an experimental study of the algorithm was conducted.

The use of single time-instance features, where entire speech utterance is used for feature computation, is not accurate and adequate in capturing the time localized information of short-time transient distortions and their distinction from plosive sounds of speech, particularly degraded by impulsive noise. Hence, the importance of estimating features at multiple time-instances is sought. In this, only active speech segments of degraded speech are used for features computation at multiple time-instances on per frame basis. Here, active speech means both voiced and unvoiced frames except silence. The features of different combinations of multiple contiguous active speech segments are computed and called multiple time-instances features. The joint GMM training has been done using these features along with the subjective MOS of the corresponding speech utterance to obtain the parameters of GMM. These parameters of GMM and multiple time-instances features of test speech are used to compute the objective MOS values of different combinations of multiple contiguous active speech segments. The overall objective MOS of the test speech utterance is obtained by assigning equal weight to the objective MOS values of the different combinations of multiple contiguous active speech segments. This algorithm outperforms the Recommendation ITU-T P.563 and recently published algorithms.

Mobile devices are widely used today for speech communication. The environments in which these devices are used are widely varied and often the level of background noise in the speaker's environment can be significant. The purpose of speech enhancement is to reduce the level of background noise, ideally to such a level that it is not noticed by the listener. While speech enhancement algorithms can significantly reduce the noise level in a speech signal, improving speech quality, it is widely recognized that enhancement algorithms can have a negative impact on speech intelligibility. This paper compares the effect of three different speech enhancement algorithms on the intelligibility and the quality of speech. This work is the initial phase of an investigation into mitigating the impact of speech enhancement algorithms on speech intelligibility. The speech enhancement algorithms evaluated each use different approaches for noise reduction, namely, a statistical model-based algorithm, a noise estimation algorithm and a wavelet packet decomposition-based algorithm. Two objective speech intelligibility measurements and three objective speech quality measurements are used to assess the performance of the enhancement algorithms. The results of the experiments show that all the speech enhancement algorithms in this study have a negative impact on speech intelligibility to varying degrees.

When speaking devices such as smartphones, tablet-PCs, or GPS systems are used in noisy outdoor environments, the intelligibility of speech significantly drops. This is even more pronounced when synthetic speech is used. This article describes how a statistical parametric speech synthesis system trained on an ordinary synthesis database can be designed to generate highly intelligible speech, even at very low signal-to-noise ratios. By using a simple and flexible vocoder based on a full-band harmonic model, the proposed system applies deterministic noise-independent modifications at several levels: speaking rate, average fundamental frequency level and range, energy contour over time, formant sharpness, and intensity of specific spectral bands. The degree of intelligibility achieved by the system has been evaluated by means of a large-scale subjective test, the results of which show that the suggested approach clearly outperforms a reference state-of-the-art TTS system and also unmodified natural speech in some conditions. In comparison with alternative systems evaluated in the same framework, the proposed one exhibits the best performance in the scenarios with lowest signal-to-noise ratio. Finally, the impact of the suggested modifications on naturalness, quality and similarity to the original natural voice is quantified by means of a subjective test.

Speech coding is the art of creating a minimally redundant representation of the speech signal that can be efficiently transmitted or stored in digital media and decoding the signal with the best possible perceptual Quality. The speech transmission in wireless networks is associated with the reduction of extra information present in signal in such a way to preserve the quality and intelligibility of speech. It is known that the lower the bit rate the lesser the quality of the reconstructed speech however there is a constant quest to achieve a better speech quality at lower bit-rates.This paper presents performance analysis for the quality of advanced hybrid speech coding techniques in Time domain, Spectral domain and perceptual domain. These analyses are implemented on three different algorithms of advanced hybrid speech coding techniques such as CELP, G729 Annex A, G723.1 to assess the quality performance for English female speaker, English male speaker and Arabic female speaker by using Mat lab simulation program. Our evaluation criterion implemented includes the following tests: Signal to Noise Ratio (SNR), Segmental Signal to Noise Ratio (SNRseg), The Log-Likelihood Ratio (LLR), The Weighted Spectral Slope (WSS), Absolute Error,Perceptual Evaluation of Speech Quality (PESQ),NONE">Rating of speechdistortion, rating of background noise and the predicted rating of overall quality.

Ultra-low-bit-rate speech coding algorithm was in great demand for many fields such as underwater speech communications. Underwater speech communication for middle-long distance has the characteristics of narrow bandwidth as well as low transmission rate, which makes the underwater speech communication much difficult. Ultra-low-bit-rate speech coding algorithm plays an important role on this occasion. More over, it will be more flexible for the underwater speech communication system if the speech coding algorithm has an embedded structure. The paper introduced the principle of an embedded speech coding algorithm with dual rates at both 300bps and 400bps based on the enhanced mixed excitation linear prediction model. The results show that this embedded ultra-low-bit-rate speech coding algorithm has satisfactory quality under both DRT and MOS test.

Here we describe a speech-synthesis method using templates that can generate recording-sentence sets for speech databases and produce natural sounding synthesized speech. Applying this method to the Japan Broadcasting Corporation (NHK) weather report radio program reduced the size of the recording-sentence set required to just a fraction of that needed by a comparable method. After integrating the recording voice of the generated recording-sentence set into the speech database, speech was produced by a voice synthesizer using templates. In a paired-comparison test, 66% of the speech samples synthesized by our system using templates were preferred to those produced by a conventional voice synthesizer. In an evaluation test using a five-point mean opinion score (MOS) scale, the speech samples synthesized by our system scored 4.97, whereas the maximum score for commercially available voice synthesizers was 3.09. In addition, we developed an automatic broadcast system for the weather report program using the speech-synthesis method and speech-rate converter. The system was evaluated using real weather data for more than 1 year, and exhibited sufficient stability and synthesized speech quality for broadcast purposes.

This paper provides an overview of instrumental models for predicting the quality of speech signals. On the basis of perceptual and cognitive characteristics of the human auditory system which are relevant for quality judgment, approaches are presented which aim at predicting overall quality, intelligibility, or other quality dimensions from measurable parameters or signal characteristics. The approaches are discussed with respect to their underlying principles, showing that perception modeling can significantly improve prediction accuracy. Application examples are presented which make use of these algorithms for offline or online prediction, adaptation, or intelligibility improvement.

This paper presents a psychoacoustic-based analysis-by-synthesis approach for compressing navigable speech sources. The approach targets multi-party teleconferencing applications, where selective reproduction of individual speech sources is desired. Based on exploiting sparsity of speech in the perceptual time-frequency domain, multiple speech signals are encoded into one mono mixture signal, which can be further compressed using a standard speech codec. Using side information indicating the active speech source for each time frequency instant enables flexible decoding and reproduction. Objective results highlight the importance of considering perception when exploiting the sparse nature of speech in the time-frequency domain. Results show that this sparsity, as measured by the preserved energy level of perceptually important time-frequency components extracted from mixtures of speech signals, is similar in both anechoic and reverberant environments. The proposed approach is applied to a series of simulated and real reverberant speech recordings, where the resulting speech mixtures are compressed using a standard speech codec operating at 32 kbps. The perceptual quality, as judged both by objective and subjective evaluations, outperforms a simple sparsity approach that does not consider perception as well as the approach that encodes each source separately. While the perceptual quality of individual speech sources is maintained, subjective tests also confirm the approach maintains the perceptual quality of the spatialized speech scene.

It is known that applying a time-frequency binary mask to very noisy speech can improve its intelligibility but results in poor perceptual quality. In this paper we propose a new approach to applying a binary mask that combines the intelligibility gains of conventional binary masking with the perceptual quality gains of a classical speech enhancer. The binary mask is not applied directly as a time-frequency gain as in most previous studies. Instead, the mask is used to supply prior information to a classical speech enhancer about the probability of speech presence in different time-frequency regions. Using an oracle ideal binary mask, we show that the proposed method results in a higher predicted quality than other methods of applying a binary mask whilst preserving the improvements in predicted intelligibility.

Narrow-band speech coding techniques were previously found to reduce the accuracy of automatic Speech Emotion Recognition (SER), as well as speech and speaker recognition rates. Artificial Bandwidth Extension (ABE) based on spectral folding and spectral envelope estimation has been applied to compressed narrowband speech to test if an improvement in SER can be achieved. The modelling and classification of speech was performed with a benchmark approach based on the GMM classifier and a set of speech acoustic parameters including MFCCs, TEO and glottal parameters. The tests used the Berlin Emotional Speech data base. In general, ABE led to an improvement of SER accuracy; however the amount of improvement varied between different features, genders, and speech compression rates. In all cases, SER accuracy with ABE was at least 10% lower than for uncompressed speech.

In this work, multiple time-scale estimates of auditory features have been introduced to capture the effect of short time-transients additive noise present over some specific active regions in a speech utterance and these multiple time-scale auditory features have been used for non-intrusive speech quality measurement. The use of single time-scale auditory features is not accurate in capturing the time localized information of short-time transient distortions and their distinction from plosive sounds of speech. Hence, the importance of estimating auditory features at multiple time-scales that is relevant for objective non-intrusive speech quality estimation. The different active speech segments obtained from voice activity detection (VAD) algorithm of speech utterance are combined across the segments in increasing number of active speech segments till all the segments of complete speech utterance are accounted for. The Lyon's auditory features of the combination of active speech segments are computed on frame by frame basis. The mean, variance, skewness and kurtosis over the frames of the auditory features are computed and concatenated to obtain multiple time-scale estimates of auditory features for the different combination active speech segments. These multiple time-scale auditory features are probabilistically modeled using Gaussian Mixture Model (GMM) to map into mean opinion score (MOS) value for each combination of active speech segments. The overall objective MOS of the degraded speech is obtained by taking average of MOS values of the combination of active speech segments. A detailed result comparison has been done with the ITU-T Recommendation P.563 for telephone band speech.

To ensure a satisfactory QoE (Quality of Experience), it is essential to establish a method that can be used to efficiently investigate recognition performance for spontaneous speech. By using this method, it is allowed to monitor the recognition performance in providing speech recognition services. It can be also used as a reliability measure in speech dialogue systems. Previously, methods for estimating the performance of noisy speech recognition based on spectral distortion measures have been proposed. Although they give an estimate of recognition performance without actually performing speech recognition, the methods cannot be applied to spontaneous speech because they require the reference speech to obtain the distortion values. To solve this problem, we propose a novel method for estimating the recognition performance of spontaneous speech with various speaking styles. The main feature is to use non-reference acoustic features that do not require the reference speech. The proposed method extracts non-reference features by openSMILE (open-Source Media Interpretation by Large feature-space Extraction) and then estimates the recognition performance by using SVR (Support Vector Regression). We confirmed the effectiveness of the proposed method by experiments using spontaneous speech data from the OGVC (On-line Gaming Voice Chat) corpus.

Majority of Speech Emotion Recognition results refer to full-band uncompressed speech signals. Potential applications of SER on various types of speech platforms pose important questions about potential effects of bandwidth limitations and compression techniques used by speech communication systems on the accuracy of SER. The current study provides answers to these questions based on SER experiments with a band-limited speech as well as compressed speech. Compression techniques included AMR, AMR-WB, AMR-WB+ and mp3 methods. The modelling and classification of speech emotions was achieved using a benchmark approach based on the GMM classifier and speech features including MFCCs, TEO and glottal time and frequency domain parameters. The tests used the Berlin Emotional Speech database with speech signals sampled at 16 kHz. The results indicated that the low frequency components (0–1 kHz) of speech as well as, the high frequency components (above 4 kHz) play an important role in SER. The mp3 compression worked better with the MFCC features than with the TEO and glottal parameters. The AMR-WB and AMR-WB+ outperformed the AMR.

A low-power and high-speed speech enhancement processor for speech enhancement of noisy inputs is proposed to realize the robust speech recognition in smart glasses. It has 3 key schemes: multi-modal speech selection, look-up table based non-linear approximation circuits, and speech detection controlled dynamic clock gating. The multi-modal speech selection scheme uses three parameters to enhance the limited accuracy of the previous uni-modal user speech selection up to 98.1%. The non-linear function approximation circuit accelerates the throughput of the speech enhancement by 10.7×. The speech detection controlled clock gating reduces the redundant power consumption by 51% when there is no user voice. The proposed speech enhancement processor achieves 1.6Msamples/s throughput and 8.3mW average power consumption with the 98.1% true positive rate of speech selection in 65nm CMOS process.

Enhancement of speech in non stationary noise environments has been a major research topic due to its wide range of applications, e.g., in hearing aids. This paper proposes a method for performing dual channel speech enhancement. The proposed method involves the estimation of speech and noise short term predictor (STP) parameters using a codebook based approach, when we have access to binaural noisy signals. The estimated STP parameters are subsequently used for enhancement in a dual channel scenario. Objective measures indicate, that the proposed method is able to improve the speech intelligibility and quality.

With the development of network and speech coding, various applications of transmitting voice over internet protocol (VoIP) come into our lives. Speech reconstruction is required, as the packet loss caused by the real-time constraints transmission protocols results in the loss of speech segments. The paper proposes a novel approach to addressing speech reconstruction problem utilizing a high-order sparse linear predictor. This predictor is able to maintain the high sparsity level involved by the classical speech model as much as possible. Then we adopt differential evolution algorithm, which is arguably one of the most powerful stochastic real-parameter optimization algorithms, to predict the sparse coefficients and missing speech segments simultaneously. Experiments show the superior performance of this new approach in both signal to noise ratio (SNR) and perceptual evaluation of speech quality (PESQ).

In this paper, the development of a speech processing pipeline on smartphones for hearing aid devices (HADs) is presented. This pipeline is used for noise suppression and speech enhancement (SE) to improve speech quality and intelligibility. The proposed method is implemented to run in real-time on Android smartphones. The results of the testing conducted indicate that the proposed method suppresses the noise and improves the perceptual quality of speech in terms of three objective measures of perceptual evaluation of speech quality (PESQ), noise attenuation level (NAL), and the coherent speech intelligibility index (CSII).

A qualified speech communication is quite important in modern communication. Thus, the implementation of security algorithms is also very important to achieve real-time applications. In this paper, we design a robust and full-duplex real-time speech communication system based on Texas Instrument's 32-bit floating point DSP TMS320C6748. We use ITU-T G.723.1 as the audio codec in this system. According to the hardware of the system and also to meet the requirement of high-quality and real-time communication, several methods are introduced in this paper to optimize the algorithm. Through the optimization, time spending in compression and decompression reduce from more than 1000ms to about 10ms which guarantees a real-time communication. Our designed system has been successfully used for the speech communication of railway system.

A voice activity detector (VAD) is a device, which analyses a speech signal and generates the signal corresponding to the period containing only noise. In the present work is offered VAD, which is intended for voice coding algorithms which primarily require correct pauses detection in human speech.

Speech disability concerns communication issues encompassing hearing, speech, language and fluency. A speech recognition system is proposed to distinguish between a normal and a person having speech disability using Mel Frequency Cepstral Coefficient. The speech disability focused in this work is stuttering which is carried out on normal female speakers and female stutter speakers. This paper presents the capability of MFCC to extract features which could be used to measure the degree of speech disability. This method is a prerequisite for designing and producing standard telecommunications equipment and services that can be used to alleviate the negative consequences of a disability.

Speech separation or enhancement algorithms seldom exploit information about phoneme identities. In this study, we propose a novel phoneme-specific speech separation method. Rather than training a single global model to enhance all the frames, we train a separate model for each phoneme to process its corresponding frames. A robust ASR system is employed to identify the phoneme identity of each frame. This way, the information from ASR systems and language models can directly influence speech separation by selecting a phoneme-specific model to use at the test stage. In addition, phoneme-specific models have fewer variations to model and do not exhibit the data imbalance problem. The improved enhancement results can in turn help recognition. Experiments on the corpus of the second CHiME speech separation and recognition challenge (task-2) demonstrate the effectiveness of this method in terms of objective measures of speech intelligibility and quality, as well as recognition performance.

In this paper, we investigate the effect of lossy speech compression on text-independent speaker verification task. We have evaluated the voice biometrics performance over several state-of-the art speech codecs including recently released Enhanced Voice Services (EVS) codec. The tests were performed in both codec-matched and codec-mismatched scenarios. The test results show that EVS outperforms other speech codecs used in our test and it can be used to generate speaker models that are quite robust to varying compression levels. It was also shown that if a speech codec of higher quality (EVS, G711) is included in training data (mismatched and partially mismatched scenarios), the automatic speaker verification (ASV) gives better results than in the case of matched scenario.

Lyon's auditory features and Multi-Resolution Auditory Model (MRAM) features for narrowband speech sampled at 8 kHz for different degradations are computed. The Gaussian Mixture Model (GMM) has been used to map these auditory features of narrowband speech into the objective mean opinion score (MOS) using Expectation Maximization (EM) algorithm. Non-intrusive speech quality assessment has been done using Lyon's auditory features and MRAM features and their efficacy has been compared in terms of the correlation coefficients between the subjective MOS and the computed objective MOS. The results in terms of root mean square error (RMSE) between the subjective MOS and the computed objective MOS are computed and compared. The results in terms of the correlation coefficients are also compared with ITU-T Recommendation P.563, a standard for non-intrusive speech quality assessment, on ITU-T supplement-23, NOIZEUS-960 and NOIZEUS-2240 databases.

Improving quality and intelligibility of speech signals in mobile devices has been studied with great interest in the past. Speech information in communication channels is usually corrupted by additive acoustic noise, reverberation or channel noise. This paper explores into the possibilities of enhancing corrupted speech using the Spectral Subtraction (SS) and Linear Prediction Coding (LPC) system for mobile applications only in acoustically noisy conditions. A Spectral Subtraction block is cascaded in series with a LPC system. For a pth order LPC system, a Levinson-Durbin based algorithm computes the LPC coefficients. Typically LPC is used as a data reduction system in speech communication but in this work, we try to find an optimum pth order LPC system that could enhance speech quality. We focus on improving speech quality and not speech intelligibility in this paper. The algorithm output will be evaluated objectively with a combined Perceptual Evaluation of Speech Quality (PESQ) and Itakura-Saito (IS) system and will be compared against Mean Opinion Scores (MOS) of various other Speech Enhancement algorithms.

This article presents a tutorial overview of models for estimating the quality experienced by users of speech transmission and communication services. Such models can be classified as either parametric or signal based. Signal-based models use input speech signals measured at the electrical or acoustic interfaces of the transmission channel. Parametric models, on the other hand, depend on signal and system parameters estimated during network planning or at run time. This tutorial describes the underlying principles as well as advantages and limitations of existing models. It also presents new developments, thus serving as a guide to an appropriate usage of the multitude of current and emerging speech quality models.

International audience; For subjective assessment of speech quality in codecs, a reference impairment system is required to introduce controlled degradations to calibrate subjective evaluation. A reference system provides a convenient means for making meaningful comparisons between subjective test results across laboratories and can be viewed as a scale on which mean opinion scores are projected, this scale being supposed to cover the whole range of quality. Nowadays, standardized anchor systems do not fit any more the degradations brought by the present codecs. This paper aims at offering new reference signals simulating the defaults of codecs currently used on telecommunication networks. Twenty wideband codecs are compared through dissimilarity tests. A multidimensional scaling technique allows us to define a four-dimensional perceptive space that appears stable for male and female talkers. A verbalization task suggests qualifying the degradations perceived by the listeners with the following attributes: muffle, background noise, noise on speech, and hiss, each conveyed by one dimension. These dimensions are correlated with objective measures such as spectral centroid, energy in the silent part in the high frequency sub-band, ratio of brightness between deterministic part and residual part of the signal and spectral correlation coefficient. New reference signals are produced and a phase of validation suggests a perceptive space quite coherent with the original one.

This paper describes a double-ended quality assessment system for speech with a bandwidth of up to 14 kHz (so-called super-wideband speech). The quality assessment system is based on a combination of local and global features, where the local features are dependent on a time alignment procedure and the global features are not. The system is evaluated over a large set of subjectively scored narrowband, wideband and super-wideband speech databases. The system performs similarly to PESQ for narrowband speech and significantly better for wideband speech.

This paper presents a novel method of enhancing esophageal speech using statistical voice conversion. Esophageal speech is one of the alternative speaking methods for laryngectomees. Although it doesn't require any external devices, generated voices usually sound unnatural compared with normal speech. To improve the intelligibility and naturalness of esophageal speech, we propose a voice conversion method from esophageal speech into normal speech. A spectral parameter and excitation parameters of target normal speech are separately estimated from a spectral parameter of the esophageal speech based on Gaussian mixture models. The experimental results demonstrate that the proposed method yields significant improvements in intelligibility and naturalness. We also apply one-to-many eigenvoice conversion to esophageal speech enhancement to make it possible to flexibly control the voice quality of enhanced speech.

This paper proposes a method of low complexity speech mixing with speech codecs based on predictive coding for multimedia conferences. The proposed method applies a filter state management (FSM) technique to a partial mixing method in order to avoid inconsistency of the filter states of encoders. The inconsistency is created by switching of the encoders when the speakers to be mixed are switched. The results of subjective evaluations of speech quality show that the proposed method avoids the inconsistency, and achieves significantly higher speech quality than the conventional partial mixing method without the FSM and almost the same speech quality as the full mixing method. The complexity evaluation results show that the proposed method achieves much lower complexity than the full mixing method.

The performance of automatic speech recognition (ASR) systems in the presence of noise is an area that has attracted a lot of research interest. Additive noise from interfering noise sources, and convolutional noise arising from transmission channel characteristics both contribute to a degradation of performance in ASR systems. This paper addresses the problem of robustness of speech recognition systems in the first of these conditions, namely additive noise. In particular, the paper examines the use of the auditory model of Li et al. [Li, Q., Soong, F.K., Siohan, O., 2000. A high-performance auditory feature for robust speech recognition. In: Proc. 6th Internat. Conf. on Spoken Language Processing (ICSLP), Vol. III. pp. 51-54] as a front-end for a HMM-based speech recognition system. The choice of this particular auditory model is motivated by the results of a previous study by Flynn and Jones [Flynn, R., Jones, E., 2006. A comparative study of auditory-based front-ends for robust speech recognition using the Aurora 2 database. In: Proc. IET Irish Signals and Systems Conf., Dublin, Ireland. pp. 111-116] in which this auditory model was found to exhibit superior performance for the task of robust speech recognition using the Aurora 2 database [Hirsch, H.G., Pearce, D., 2000. The Aurora experimental framework for the performance evaluation of speech recognition systems under noisy conditions. In: Proc. ISCA ITRW ASR2000, Paris, France. pp. 181-188]. In the speech recognition system described here, the input speech is pre-processed using an algorithm for speech enhancement. A number of different methods for the enhancement of speech, combined with the auditory front-end of Li et al., are evaluated for the purpose of robust connected digit recognition. The ETSI basic [ETSI ES 201 108 Ver. 1.1.3, 2003. Speech processing, transmission and quality aspects (STQ); distributed speech recognition; front-end feature extraction algorithm; compression algorithms] and advanced [ETSI ES 202 050 Ver. 1.1.5, 2007. Speech processing, transmission and quality aspects (STQ); distributed speech recognition; advanced front-end feature extraction algorithm; compression algorithms] front-ends proposed for DSR are used as a baseline for comparison. In addition to their effects on speech recognition performance, the speech enhancement algorithms are also assessed using perceptual speech quality tests, in order to examine if a correlation exists between perceived speech quality and recognition performance. Results indicate that the combination of speech enhancement pre-processing and the auditory model front-end provides an improvement in recognition performance in noisy conditions over the ETSI front-ends.

We present a method for the objective quality evaluation of noise-reduced speech in wideband speech communication services, which utilize speech with a wider bandwidth (e. g., 7kHz) than the usual telephone bandwidth. Experiments indicate that the amount of residual noise and the distortion of speech and noise, which are quality factors, influence the perceived quality degradation of noise-reduced speech. From the results, we observe the principal relationships between these quality factors and perceived speech quality. On the basis of these relationships, we propose a method that quantifies each quality factor in noise-reduced speech by analyzing signals that can be measured and assesses the over-all perceived quality of noise-reduced speech using values of these quality factors. To verify the validity of the method, we perform a subjective listening test and compare subjective quality of noise-reduced speech with its estimation. In the test, we use various types of background noise and noise-reduction algorithms. The verification results indicate that the correlation between subjective quality and its objective estimation is sufficiently high regardless of the type of background noise and noise-reduction algorithm.

In this paper, we propose a multi-stage speech enhancement technique for speech recognition. At first, a multi-channel speech enhancement method takes advantage of the spatial information of speech source. Then, in the second stage, single-channel speech enhancement based on data-driven approach is adopted to improve performance of speech recognition at server side. This method can improve the quality of speech signal which maximizes the advantage of each speech enhancement technique. The experimental result shows that the proposed technique is superior to conventional multi-stage speech enhancement algorithms.

Existing standard speech coders can provide high quality speech communication. However, they tend to degrade the performance of automatic speech recognition (ASR) systems that use the reconstructed speech. The main cause of the degradation is in that the linear predictive coefficients (LPCs), which are typical spectral envelope parameters in speech coding, are optimized to speech quality rather than to the performance of speech recognition. In this paper, we propose a speech coder using mel-frequency cepstral coefficients (MFCCs) instead of LPCs to improve the performance of a server-based speech recognition system in network environments. To develop the proposed speech coder with a low-bit rate, we first explore the interframe correlation of MFCCs, which results in the predictive quantization of MFCC. Second, a safety-net scheme is proposed to make the MFCC-based speech coder robust to channel errors. As a result, we propose an 8.7 kbps MFCC-based CELP coder. It is shown that the proposed speech coder has a comparable speech quality to 8 kbps G.729 and the ASR system using the proposed speech coder gives the relative word error rate reduction by 6.8% as compared to the ASR system using G.729 on a large vocabulary task (AURORA4).

Methods for speech compression aim at reducing the transmission bit rate while preserving the quality and intelligibility of speech. These objectives are antipodal in nature since higher compression presupposes preserving less information about the original speech signal. This paper presents a method for compressing speech based on polynomial approximations of the trajectories in time of various speech features (i.e., spectrum, gain, and pitch). The compression method can be integrated into frame-based speech coders, and can also be applied to features that can be represented as temporal series greater in duration than the frame interval. Theoretical issues and experimental results regarding this type of compression are addressed in this paper. Experimental implementation into a 2400 b/s standard speech coder is reported along with objective and subjective evaluations of operation in various noise environments. The new speech coder operates at a transmission rate of 1533 b/s, and for all noisy conditions tested performs better than the 2400 b/s standard speech coder

The use of text to speech (TTS) systems beyond pure scientific applications has been gaining ground and has been incorporated in everyday life aspects. Thus, the development of TTS system, that can produce synthesized speech of high quality is more than necessary in state-of-the-art everyday life applications. This paper evaluates the performance of a TTS using harmonic plus noise model (HNM) - a speech analysis/synthesis algorithm, as part of its DSP module. A TTS for the Greek language is studied with respect to its DSP module and an evaluation is presented of the HNM algorithm in terms of overall performance and the impact of its attributes on the results quality.

At low bit-rates, speech coders relying on a single source model have difficulties to properly render speech and background noise simultaneously. Difficulties can get even bigger when using speech enhancement techniques within the coding scheme: these have shown to improve quality for clean speech, but introduce unpleasant instabilities under noisy conditions. This paper presents a novel approach named Comfort Noise Addition (CNA) for post-processing noisy speech coding. It continuously injects comfort noise in both active and inactive segments of the decoded speech. Based on an accurate noise estimate, it enhances the reproduction of the background noise while masking artefacts from the coding and the speech enhancement. Listening tests confirm that CNA allows speech coders to compensate for the limitations of speech enhancement and to significantly improve the quality of noisy speech at low bit-rates. CNA was adopted in the recent 3GPP codec for Enhanced Voice Services (EVS).

The advanced front-end (AFE) for automatic speech recognition (ASR) was standardized by the European Telecommunications Standards Institute (ETSI). The AFE provides speech enhancement realized by an iterative Wiener filter (IWF) in which a smoothed FFT spectrum over adjacent frames is used to design the filter. We have previously proposed robust time-varying complex AR (TV-CAR) speech analysis and evaluated the performance of speech processing such as F0 estimation and speech enhancement. TV-CAR analysis can estimate more accurate spectrum than FFT, especially in low frequencies because of the nature of the analytic signal. In addition, the TV-CAR can estimate more accurate speech spectrum against additive noise. In this paper, the time-invariant version of wide-band TV-CAR analysis is introduced to the IWF in the AFE and is evaluated using the CENSREC-2 database.

Voice activity detection of speech signal is an important part in the process of speech recognition. The accuracy of voice activity detection is directly related to the accuracy of speech recognition. In this paper, the use of locomotive voice as test data, using the traditional dual-threshold method for voice activity detection. It is found that under the quiet and noisy conditions the traditional dual-threshold method obtained information of speech voice activity detection having lots of errors. In view of the above problems, this paper presented an improved dual-threshold method to make voice activity detection, processing speech signal, short-time average energy and zero crossing rate. And through MATLAB simulation, the experimental results show that under the quiet and noisy conditions comparing the improved method proposed by this paper and the traditional method, are closer to real speech voice activity part in the test data.

Speech quality evaluation is a very complex task with important applications. With the deployment of 3G and 4G networks, the end-user is requiring more quality of service. Currently, the assessment of speech quality over mobile phones is done in a controlled environment with a reference clean signal and its distorted version. However, this does not evaluate the quality of speech in a real-time situation as it demands a reference signal. In this paper, it is introduced a new method to classify a speech signal over a mobile device as distorted or not. The degraded signal is evaluated and its quality is scored similarly as what is done by PESQ (Perceptual Evaluation of Speech Quality). The experiments proved that the method is quite efficient with an average difference of 12% comparing to PESQ scores which is based on a reference clean signal.

In this paper a multi-channel speech enhancement framework for distant speech acquisition in noisy and reverberant environments for Non-negative Matrix Factorization (NMF)-based Automatic Speech Recognition (ASR) is proposed. The system is evaluated for its use in an assistive vocal interface for physically impaired and speech-impaired users. The framework utilises the Spatially Pre-processed Speech Distortion Weighted Multi-channel Wiener Filter (SP-SDW-MWF) in combination with a postfilter to reduce noise and reverberation. Additionally, the estimation uncertainty of the speech enhancement framework is propagated through the Mel-Frequency Cepstrum Coefficients (MFCC) feature extraction to allow for feature compensation in a later stage. Results indicate that a) using a trade-off parameter between noise reduction and speech distortion has a positive effect on the recognition performance with respect to the well-known GSC and MWF and b) the addition of a postfilter and the feature compensation increases performance with respect to several baselines for a non-pathological and pathological speaker. Dekkers G., van Waterschoot T., Vanrumste B., Van Den Broeck B., Gemmeke J.F., Van hamme H., Karsmakers P., ''A multi-channel speech enhancement framework for robust NMF-based speech recognition for speech-impaired users'', Proceedings 16th annual conference of the International Speech Communication Association (ISCA) - Interspeech 2015, pp. 746-750, September 6-10, 2015, Dresden, Germany. ispartof: pages:746-750 ispartof: INTERSPEECH 2015: proceedings vol:2015-January pages:746-750 ispartof: INTERSPEECH 2015 location:Dresden, Germany date:6 Sep - 10 Sep 2015 status: published

Automatic recognition of emotions in speech has attracted the attention of the research community in recent years. Some of the most relevant proposed applications of it are in call-centers. In these scenarios the speech is distorted by compression algorithms. The effects of such distortion on the performance of systems for automatic recognition of emotions must be assessed. In this study these effects are evaluated independently of any other distortions generated by the communications channel. Several state-of-the-art codecs are used to compress the speech signals of two emotional speech databases. The databases used are the Berlin Database of Emotional Speech and the enterface05. The methodology considers voiced and unvoiced segments of the speech separately. Spectral, cepstral, noise and Non-Linear Dynamics (NLD) measures are used to characterize the segments. Finally, a classifier based on a Gaussian Mixture Model (GMM) is used to identify the emotion. The results indicate that voiced segments are less affected by the compression than unvoiced ones in terms in classification accuracy. They also show that the bandwidth of the analyzed signals is an important factor in the classification results.

The majority of speech coding algorithms are based on the code excited linear prediction (CELP) paradigm, modelling the speech signal by linear prediction. This coding approach offers the advantage of a very short algorithmic delay, due to the windowing scheme based on rectangular windowing of the residual of the linear predictor. Although widely used, the performance and structural choices of this windowing scheme have not been extensively documented. In this paper we introduce three alternative windowing schemes, as alternatives to the one already used in CELP codecs. These windowing schemes differ in their handling of transitions between frames. Our subject evaluation shows that omitting the error feedback loop yields an increase in perceptual quality at scenarios with high quantization noise. In addition, objective measures show that while error feedback improves the accuracy slightly at high bitrates, at low bitrates it causes a degradation in quality, resulting in a lower SNR.

This plenary session will cover speech processing research advances with the emphasis on speech and audio coding methods. In the session, we will discuss the fundamental principles, techniques, and algorithms used in current coding applications including a summary of codecs for telecommunication standards. The session will start with a discussion on: the basic speech representation methods, the performance measures used to evaluate coded speech, and the role of the standards. Brief algorithm descriptions include: ADPCM, sub-band coding, adaptive transform coding, sinusoidal transform coding (STC), linear predictive coding (LPC), and analysis-by-synthesis LPC (sparse excitation, code excited LPC, and ACELP). The presentation will feature audio, and computer demonstrations of recent speech coding standards including voice-over IP algorithms. The plenary session will also cover wideband audio standards such as MPEG audio and other layers (e.g., MP3, AAC). Recent algorithms will also be described including the following: Variable-Rate Multimode Wideband (VMR-WB), Speex, G722.1, OGG Vorbis 2012, iLBC, SELT, SILK, Opus 2013, Qualcomm wideband 5G codecs. At the end of the session, we will cover briefly recent applications that use voice features for detecting speech pathologies, and also discuss how long-term speech parameters can be used as predictors of other diseases such as tremors, Alzheimer's etc.

Objective evaluation of speech quality is an important part of the quality of communications service, and the speech in high-noise environments affect the people's auditory perception, which is an important factor for people to determine the speech quality. In this paper, a non-intrusive evaluation method for speech quality in high-noise environments is proposed, and a noise tracking and subtraction algorithm is employed to obtain the quasi clean speech from the high noisy speech. The quasi clean speech is regarded as the reference speech, and a perceptual model, which is a modified version of ITU-T PESQ, is applied to acquire the Mean Opinion Score (MOS) via the measurement of the distortions between the quasi clean speech and the high noisy speech. In the proposed model, only the noisy speech is needed for speech quality evaluation, and the quasi clean speech construction transforms non-intrusive estimation to PESQ intrusive measure. Experimental results demonstrate that the correlation coefficient of the proposed method is 0.906 on NOIZEUS dataset, and is 4.3% superior to ITU-T standard recommendation P.563, which is the state of the art non-intrusive method.

In this paper, an adaptive speech streaming method is proposed to improve the perceived speech quality (PSQ) of voice over wireless multimedia sensor network (WMSNs). First of all, the proposed method estimates the PSQ of the received speech data under different network conditions that are represented by the packet loss rates (PLRs). Simultaneously, the proposed method classifies the speech signal as either an onset or a nononset frame. Based on the estimated PSQ and the speech class, it determines an appropriate bit rate for the redundant speech data (RSD) that are transmitted with the primary speech data (PSD) to help reconstruct the speech signals of any lost frames. In particular, when the estimated PLR is high, the bit rate of the RSD should be increased by decreasing that of the PSD. Thus, the bandwidth of the PSD is changed from wideband to narrowband, and an artificial bandwidth extension technique is applied to the decoded narrowband speech. It is shown from the simulation that the proposed method significantly improves the decoded speech quality under packet loss conditions in a WMSN, compared to a decoder-based packet loss concealment method and a conventional redundant speech transmission method.

To approximate the speech quality of a given speech enhancement system, most of the existing instrumental metrics rely on the calculation of a distortion metric defined between the clean reference signal and the enhanced signal in the spectral amplitude domain. Several recent studies have demonstrated the effectiveness of employing a phase modification stage in single-channel speech enhancement showing positive impact brought by modifying both amplitude and phase in contrast to the conventional methods where the noisy spectral amplitude is only modified and noisy phase is used for signal reconstruction. In this work we present two contributions; First we study the reliability of the existing instrumental metrics for performance evaluation of phase-aware methods, and second we propose novel phase-aware instrumental metrics and evaluate their reliability in terms of predicting the perceived quality achieved by the phase-aware methods. Our objective and subjective evaluations demonstrate that PESQ and the proposed phase deviation metric perform as reliable speech quality estimators following the subjective results.

Objective speech quality assessment is done to replace the time taking and cumbersome subjective listening test to assess the quality of degraded speech processed by different speech processing algorithms. For performance evaluation, all objective speech quality assessment algorithms require the Mean Opinion Score-Listening Quality Subjective (MOS-LQS) or subjective MOS obtained from the subjective listening test for comparison with the Mean Opinion Score-Listening Quality Objective (MOS-LQO) or objective MOS. This paper gives details of the subjective database created for the different applications. In this work, subjective speech quality assessment has been done for speech utterances at different degradation/noise conditions to assign the MOS-LQS or subjective MOS and compared with MOS-LQO or the objective MOS obtained from an algorithm based on relevant auditory perception features for an illustration. Two different databases NOIZEUS-960 and NOIZEUS-2240 having 3200 speech utterances processed by different speech enhancement algorithms at different types of real life noise conditions have been used for subjective listening test by real life users. The statistical analysis of the MOS-LQS assigned by different subjects for different degraded speech utterances has been done and its details are reported in the paper. Further, the correlation coefficients are computed between the subjective MOS and the objective MOS for both the databases for an illustration and inferences are drawn regarding the usefulness of the subjective MOS in speech quality assessment algorithms.