Residual Mel Frequency Cepstral Coefficient Research Articles

Detection of replay signals using excitation source and shifted CQCC features

The replay attack is refereed as an unauthorized attempt to access the automatic speaker verification (ASV) system by using the pre-recorded speech samples of any target. The replay attack is performed by placing the pre-recorded speech sample of the target before the machine. Of late the replay attack is identified as the greatest threat to ASV system, mainly due to the availability of high quality recording and playback devices. In this work, excitation source feature referred as glottal mel frequency cepstral coefficient (GMFCC) and shifted constant Q cepstral coefficient (SCQCC) are proposed for detection of replay signals. The GMFCC is derived by applying conventional mel-cepstral technique to glottal flow derivative signal. The SCQCC is computed by using constant Q cepstral processing. The effectiveness of the proposed features are demonstrated by conducting experiments with ASVspoof 2017 version 2.0 database. The proposed GMFCC feature provides an equal error rate (EER) of 16.78%, that is 19.63% higher than the recently proposed residual mel frequency cepstral coefficient(RMFCC) feature. The conventional CQCC feature provides an EER of 12.32%. The proposed SCQCC feature provides an EER of 11.34%, shows a relative improvement of 7.94% over CQCC. Further, the CQCC in together with proposed GMFCC provides an EER of 8.82%. On the other hand, the proposed SCQCC+GMFCC system provides an EER of 8.60%. These results signify the usefulness of the proposed system to counter replay attacks.

Exploration of excitation source information for shouted and normal speech classification.

Discrimination between shouted and normal speech is an essential prerequisite for many speech processing applications. Existing works have established that excitation source information plays a significant role in shouted speech production. In speech processing literature, various features have been proposed to model different aspects of the excitation source. The principal contribution of this work is to explore three such features, Discrete Cosine Transform of Integrated Linear Prediction Residual (DCT-ILPR), Mel-Power Difference of Spectrum in Sub-bands (MPDSS), and Residual Mel-Frequency Cepstral Coefficient (RMFCC), for shouted and normal speech classification. The DCT-ILPR feature represents the shape of the glottal cycle, MPDSS estimates the periodicity of the excitation source spectrum, and RMFCC characterizes smoothed spectral information of the excitation source. The authors have also contributed a dataset containing shouted and normal speech. This work is evaluated on three datasets and benchmarked against three baseline methods. Deep neural networks are used to study the classification performance of individual features and their combinations. The generalization performance of features (and combinations) is also investigated. Fusion of excitation source features with Mel-Frequency Cepstral Coefficients (MFCC) provides the best performance compared to other combinations. Noise analysis shows that adding excitation features with MFCC+ΔΔ provides a more robust classification system.

Usefulness of linear prediction residual for replay attack detection

This work demonstrates the usefulness of processing linear prediction (LP) residual signal for detecting replay attacks. The playback device having non-flat frequency response modulates the input signal passing through it, resulting spectral distortion in replay signals especially in the low frequency regions (0–300 Hz). In effect, the excitation source information present below 300 Hz gets distorted. The linear prediction (LP) residual signal implicitly contains excitation source information. The excitation source feature, residual mel frequency cepstral coefficients (RMFCC) obtained from the LP residual signal has been proposed for replay detection task. The significance of RMFCC feature has been investigated through Gaussian mixtures model-universal background model (GMM-UBM) ASV experiments and spoof detection experiments using self-developed IITG-MV replay database and standard ASVspoof 2017 database, respectively. For IITG-MV, relative tandem-detection cost function (t-DCF) improvements of 29.78% (male), 7.96% (female) and 24.52% (whole-set) are observed for RMFCC+MFCC combination over MFCC feature. For ASVspoof 2017, relative EER improvements of 37.34% are reported for RMFCC+CQCC combination over CQCC feature. Minimum reported EER is 9.50% for RMFCC+CQCC system. Experimental results shows that the RMFCC feature contains information complementary to MFCC and CQCC features.

Combining evidences from Hilbert envelope and residual phase for detecting replay attacks

In this work, the Hilbert envelope of the linear prediction (LP) residual and the residual phase have been explored for detecting replay attacks. The two source features namely, LP residual Hilbert envelope mel frequency cepstral coefficient (LPRHEMFCC) and residual phase cepstral coefficient (RPCC) are used for replay detection. From the signal perspectives, Hilbert envelope represents the amplitude information of LP residual samples. Residual phase represents to excitation information present in the sequence of LP residual samples. Hence, both can be considered as two components of the raw LP residual signal. In this direction, score level fusion of LPRHEMFCC and RPCC features is compared with a third source feature named as, residual mel frequency cepstral coefficient (RMFCC) derived from the raw LP residual using LP analysis. Comparative analysis has been performed using Gaussian mixtures model-universal background model (GMM-UBM) ASV experiments (IITG-MV replay database) and spoof detection experiments (ASVspoof 2017 database). For IITG-MV database, relative (RFAR-ZFAR) improvements of 86.10% (males), 27.45% (females) and 54.14% (whole-set) are achieved for (LPRHEMFCC + RPCC) + MFCC combination over RMFCC + MFCC combination. The RFAR and ZFAR stands for false acceptance rate under replay attacks and zero effort impostor attacks, respectively. In terms of tandem-detection cost function (t-DCF) metrics, the obtained relative improvements are 40.50%, 13.13% and 26.16%, respectively. For ASVspoof 2017 database, relative EER improvements of 11.72% and 6.74% are achieved for (LPRHEMFCC + RPCC) + MFCC and (LPRHEMFCC + RPCC) + CQCC over RMFCC + MFCC and RMFCC + CQCC, respectively. These observations justify the usefulness of exploring Hilbert envelope and residual phase components of the LP residual over direct processing of the LP residual signal for detecting replay attacks. Moreover, score level fusion of LPRHEMFCC, RPCC and CQCC provides 8.86% EER.

Exploring Text-Constraint Models and Source Information for Long-Enrollment with Short-Test Speaker Verification

This work focuses on long-enrollment with short-test speaker verification (SV) from the perspective of application-oriented systems. The importance of phonetic match between train and test models is explored in terms of having a text-constraint model-based framework on Part IV of RedDots database. This database has a text-dependent and a text-prompted-based enrollment conditions for speaker modeling. Two different text-constraint setups are formalized for evaluating the effect of text match on train and test sessions. Further, the excitation source features mel power difference of spectrum in subbands, residual mel frequency cepstral coefficient and discrete cosine transform of integrated linear prediction residual are investigated to determine their significance for text-constraint-based framework. Although the source features individually perform poorer compared to the conventional mel frequency cepstral coefficient (MFCC) features, their significance is reflected in fusion due to the complementary nature of information carried by them. Additionally, the source features become imperative for text-constraint-based models for long-enrollment with short-test SV in fusion to MFCC features and achieves commendable improvement from baseline framework of text-prompted-based enrollment condition. This thus minimizes the performance difference between text-dependent and text-prompted-based enrollment condition showing importance of text-constraint models and source information in long-enrollment with short-test-based framework favorable from the perspective of field deployable systems.

Improvement of phone recognition accuracy using speech mode classification

In this work, we have developed a speech mode classification model for improving the performance of phone recognition system (PRS). In this paper, we have explored vocal tract system, excitation source and prosodic features for development of speech mode classification (SMC) model. These features are extracted from voiced regions of a speech signal. In this study, conversation, extempore, and read speech are considered as three different modes of speech. The vocal tract component of speech is extracted using Mel-frequency cepstral coefficients (MFCCs). The excitation source features are captured through Mel power differences of spectrum in sub-bands (MPDSS) and residual Mel-frequency cepstral coefficients (RMFCCs) of the speech signal. The prosody information is extracted from pitch and intensity. Speech mode classification models are developed using above described features independently, and in fusion. The experiments carried out on Bengali speech corpus to analyze the accuracy of the speech mode classification model using the artificial neural network (ANN), naive Bayes, support vector machines (SVMs) and k-nearest neighbor (KNN). We proposed four classification models which are combined using maximum voting approach for optimal performance. From the results, it is observed that speech mode classification model developed using the fusion of vocal tract system, excitation source and prosodic features of speech, yields the best performance of 98%. Finally, the proposed speech mode classifier is integrated to the PRS, and the accuracy of phone recognition system is observed to be improved by 11.08%.

Exploring different attributes of source information for speaker verification with limited test data.

This work explores mel power difference of spectrum in subband, residual mel frequency cepstral coefficient, and discrete cosine transform of the integrated linear prediction residual for speaker verification under limited test data conditions. These three source features are found to capture different attributes of source information, namely, periodicity, smoothed spectrum information, and shape of the glottal signal, respectively. On the NIST SRE 2003 database, the proposed combination of the three source features performs better [equal error rate (EER): 20.19%, decision cost function (DCF): 0.3759] than the mel frequency cepstral coefficient feature (EER: 22.31%, DCF: 0.4128) for 2 s duration of test segments.

Processing of linear prediction residual in spectral and cepstral domains for speaker information

In this work the linear prediction (LP) residual is processed in spectral and cepstral domains to model the speaker-specific excitation information. In the spectral domain, the excitation energy information is modeled from subband energies (SBE). The excitation periodicity information is modeled by power differences of spectrum in subband (PDSS) measure. This work carries some refinements in the existing methods of extracting SBE and PDSS by exploiting the nature of the excitation spectrum. The SBE and PDSS values are computed from mel warped residual subband spectrum and called as residual mel subband energies (R-MSE) and mel power differences of subband spectra (M-PDSS), respectively. The different speaker recognition studies performed using NIST-99 and NIST-03 databases demonstrate that R-MSE and M-PDSS features represent good speaker information. It is also demonstrated that the excitation energy information can be better modeled in the cepstral domain by residual mel frequency cepstral coefficients (R-MFCC). Furhter, the evidences provided by M-PDSS and R-MFCC features are different and combine well and provides improved recognition performance. The combined evidence from M-PDSS and R-MFCC together with the vocal tract information further improves the performance. Finally, a comparative study on processing the LP residual in temporal, spectral and cepstral domains demonstrates that with a small compromise with the recognition performance, processing LP residual in spectral and cepstral domains provide compact and effective way of representing the excitation information, as compared to temporal processing.

Speaker Recognition using Supra-segmental Level Excitation Information

Speaker specific information present in the excitation signal is mostly viewed from sub-segmental, segmental and supra-segmental levels. In this work, the supra-segmental level information is explored for recognizing speakers. Earlier study has shown that, combined use of pitch and epoch strength vectors provides useful supra-segmental information. However, the speaker recognition accuracy achieved by supra-segmental level feature is relatively poor than other levels source information. May be the modulation information present at the supra-segmental level of the excitation signal is not manifested properly in pith and epoch strength vectors. We propose a method to model the supra-segmental level modulation information from residual mel frequency cepstral coefficient (R-MFCC) trajectories. The evidences from R-MFCC trajectories combined with pitch and epoch strength vectors are proposed to represent supra-segmental information. Experimental results show that compared to pitch and epoch strength vectors, the proposed approach provides relatively improved performance. Further, the proposed supra-segmental level information is relatively more complimentary to other levels information.