Empirical Mode Decomposition for adaptive AM-FM analysis of Speech

Mel-frequency cepstral coefficients have long been the most widely used type of speech representation. They were introduced to incorporate biologically inspired characteristics into artificial speech recognizers. Recently, the introduction of new alternatives to the classic mel-scaled filterbank has led to improvements in the performance of phoneme recognition in adverse conditions. In this work we propose a new bioinspired approach for the optimization of the filterbanks, in order to find a robust speech representation. Our approach--which relies on evolutionary algorithms--reduces the number of parameters to optimize by using spline functions to shape the filterbanks. The success rates of a phoneme classifier based on hidden Markov models are used as the fitness measure, evaluated over the well-known TIMIT database. The results show that the proposed method is able to find optimized filterbanks for phoneme recognition, which significantly increases the robustness in adverse conditions.

/pdf/evolutionary-splines-for-cepstral-filterbank-optimization-in-2hwwny2h95.pdf

Evolutionary splines for cepstral filterbank optimization in phoneme classification

Evolutionary algorithms provide flexibility and robustness required to find satisfactory solutions in complex search spaces. This is why they are successfully applied for solving real engineering problems. In this work we propose an algorithm to evolve a robust speech representation, using a dynamic data selection method for reducing the computational cost of the fitness computation while improving the generalisation capabilities. The most commonly used speech representation are the mel-frequency cepstral coefficients, which incorporate biologically inspired characteristics into artificial recognizers. Recent advances have been made with the introduction of alternatives to the classic mel scaled filterbank, improving the phoneme recognition performance in adverse conditions. In order to find an optimal filterbank, filter parameters such as the central and side frequencies are optimised. A hidden Markov model is used as the classifier for the evaluation of the fitness for each individual. Experiments were conducted using real and synthetic phoneme databases, considering different additive noise levels. Classification results show that the method accomplishes the task of finding an optimised filterbank for phoneme recognition, which provides robustness in adverse conditions.

http://people.uwplatt.edu/~borilh/citing_papers/Vignolo_Evolutionary%20Cepstral%20Coefficients.pdf

Evolutionary cepstral coefficients

Analysis of the Hilbert Spectrum for Text-Dependent Speaker Verification

Feature optimisation for stress recognition in speech

Mel-frequency cepstral coefficients (MFCC) are the most widely used features for speech recognition. However, MFCC-based speech recognition performance degrades in presence of additive noise. In this paper, we propose a set of noise-robust features based on conventional MFCC feature extraction method. Our proposed method consists of two steps. In the first step, mel sub-band Wiener filtering is carried out. The second step consists of estimating SNR in each sub-band and calculating the sub-band entropy by defining a weight parameter based on sub-band SNR to entropy ratio. The weighting has been carried out in a way that gives more important roles, in cepstrum parameter formation, to sub-bands that are less affected by noise. Experimental results indicate that this method leads to improved ASR performance in noisy environments. Furthermore, due to the simplicity of the implementation of our method, its computational overhead in comparison to MFCC is quite small.

Weighting of Mel Sub-bands Based on SNR/Entropy for Robust ASR

In this paper, we present a new back-end classifier for GMM-LM based language identification systems. Our new proposed system consists of two main parts, mapping matrix and bank of SVMs. These two parts are located in series after GMM-LM system. The mapping matrix, maps the language models' output vectors to a new space in which the languages are more separable than before. Then each SVM in the SVM bank separates one language from the others. We used a new sequence kernel for each SVM in the bank. We show that our new sequence kernel-based SVMs separate languages more efficiently than common Gaussian mixture and GLDS SVM back-end classifiers. Also our new mapping matrix outperforms common linear discriminant matrix in separating classes from each other. Using these two parts increases the LID accuracy noticeably in comparison with the other LDA-GMM and LDA-GLDS SVM back-end classifiers. Our experiments on 5 languages from OGI-TS multilanguage task, prove our claim.

Spoken Language Identification Using a New Sequence Kernel-based SVM Back-end Classifier

The duty of speaker diarization comprises of answering the question ldquoWho spoke when?rdquo. In this paper, we present an approach comprising of PSO (particle swarm optimization) algorithm, which encodes possible segmentations of an audio record by measuring mutual information between these segments and the audio data.. This measure is used as the fitness function for the PSO. This algorithm has been tested on two actual sets of data with up to 8 speakers for speaker diarization, and has led to very good results in all test problems. The results have been compared to the same approach using genetic algorithm (GA) and the widely used DISTBIC algorithm in several practical situations, and found to be superior in most of the cases. No assumptions have been made about prior knowledge of speech signal characteristics. However, we assume that the speakers do not speak simultaneously and that we have no real-time constraints.

Speaker diarization in a multi-speaker environment using particle swarm optimization and mutual information

This paper shows research performed into the topic of speaker diarization for multi-speaker environment. It looks into the algorithms and the implementation of an offline speaker segmentation and indexing system for recorded speech data where usually more than one speaker is present. Speaker diarization is a well studied topic in the domain of broadcast news recordings. Most of the proposed systems involve hierarchical clustering of the data, where the number of speakers and their identities are known a priori. Speaker diarization is the task of assigning a unique label to all speech segments in an audio stream by the same speaker. There are two key challenges: processing speed and robustness in the presence of noise. In this paper we address the robustness issue by using a method already successful in speech recognition application. Using ANS (Autocorrelation-Based Noise Subtraction) for robust genetic algorithm-based speaker diarization, we compare the results with the baseline MFCC-based system in clean and noisy conditions.

Robust Speaker Diarization in a Multi-Speaker Environment Using Autocorrelation-based Noise Subtraction

cepstral coefficients (MFCC)arethe robustness comparedto conventional full-band ones, mostwidely usedfeatures forspeech recognition. However,particularly inthepresence offrequency selective noise [10]MFCC-based speech recognition performance degrades in [13]. Thisisduetotheir ability toisolate noise corruption presence ofadditive noise. Inthispaper, we propose a setof within particular frequency-bands bysplitting thefrequency noise-robust features basedon conventional MFCC featurespectrum intosub-bands. Sub-bandspeech recognition extraction method. Ourproposed methodconsists oftwosteps. Inthefirst step, melsub-band Wiener filtering iscarried out.systems usually extract features fromindividual sub-bands Thesecond step consists ofestimating SNRineachsub-band and basedontheassumption thatsub-band datastreams are calculating thesub-band entropy bydefining aweight parameterindependent. based onsub-band SNRtoentropy ratio. Theweighting hasbeen Conventional mel-frequency cepstral coefficients (MFCC) carried outinawaythat gives moreimportant roles, incepstrumshowgoodperformance inclean speech recognition. Inspite parameter formation, tosub-bands thatareless affected by oftheir popularity, theyhaveshownpoorperformance in noise. Experimental results indicate thatthis methodleads to noisy conditions. Ontheother hand, MFCCisaneasyandfast improvedASR performancein noisyenvironments, front-end toimplement. Hence, methods thatcanimprove Furthermore, duetothesimplicity oftheimplementation ofour MFCC features' robustness arevaluable, suchasrootcepstral method, itscomputational overhead incomparison toMFCC is analysis [13] whichsubstitutes logfunction withconstant root function andyields rootMFCC.In[12] thelog function is Robustspeechrecognition, Sub-bandWiener substituted bya variable rootfunction. Rootfunction has filtering, Melfrequency cepstral coefficients. shownbetter performance incomparison tologfunction. To overcome MFCCproblems andalso useMFCC simplicity, we propose aframework tocompensate additive noise effects on I.INTRODUCTION MFCC feature extraction. Ourmethodincludes twosteps. Speechrecognitionperfo.ance degradesdramaticallyw Initially, weapply sub-band Wiener filtering inmelsub-bands. Speech recognition performance degrades dramatically when Thntecnetoa'FCfaue r on sn speech~~~~~~~~ ~ is corpe by bakrudnis,nhne Then, theconventional MFCC features arefoundusing a

S.M. Mirrezaie

Papers

Weighting of Mel Sub-bands Based on SNR/Entropy for Robust ASR

Spoken Language Identification Using a New Sequence Kernel-based SVM Back-end Classifier

Speaker diarization in a multi-speaker environment using particle swarm optimization and mutual information

Robust Speaker Diarization in a Multi-Speaker Environment Using Autocorrelation-based Noise Subtraction

Weighting ofMelSub-bands BasedonSNR/Entropy