Relative Spectral Perceptual Linear Prediction Research Articles

Improved Multi-Model Classification Technique for Sound Event Detection in Urban Environments

Sound event detection (SED) plays an important role in understanding the sounds in different environments. Recent studies on standardized datasets have shown the growing interest of the scientific community in the SED problem, however, these did not pay sufficient attention to the detection of artificial and natural sound. In order to tackle this issue, the present article uses different features in combination for detection of machine-generated and natural sounds. In this article, we trained and compared a Stacked Convolutional Recurrent Neural Network (S-CRNN), a Convolutional Recurrent Neural Network (CRNN), and an Artificial Neural Network Classifier (ANN) using the DCASE 2017 Task-3 dataset. Relative spectral–perceptual linear prediction (RASTA-PLP) and Mel-frequency cepstrum (MFCC) features are used as input to the proposed multi-model. The performance of monaural and binaural approaches provided to the classifier as an input is compared. In our proposed S-CRNN model, we classified the sound events in the dataset into two sub-classes. When compared with the baseline model, our obtained results show that the PLP-based ANN classifier improves the individual error rate (ER) for each sound event, e.g., the error rate (ER) is improved to 0.23 for heavy vehicle events and 0.32 for people walking, and minor gains are shown in other events as compared to the baseline. Our proposed CRNN performs well when compare to the baseline and to our proposed ANN model. Moreover, in cross-validation trials, the results in the evaluation stage demonstrate a significant improvement compared to the best performance of DCASE 2017 Task-3, reducing the ER to 0.11 and increasing the F1-score by 10% in the evaluation dataset. Erosion and dilation were used during post-processing.

Developing sequentially trained robust Punjabi speech recognition system under matched and mismatched conditions

Development of a native language robust ASR framework is very challenging as well as an active area of research. Although an urge for investigation of effective front-end as well as back-end approaches are required for tackling environment differences, large training complexity and inter-speaker variability in achieving success of a recognition system. In this paper, four front-end approaches: mel-frequency cepstral coefficients (MFCC), Gammatone frequency cepstral coefficients (GFCC), relative spectral-perceptual linear prediction (RASTA-PLP) and power-normalized cepstral coefficients (PNCC) have been investigated to generate unique and robust feature vectors at different SNR values. Furthermore, to handle the large training data complexity, parameter optimization has been performed with sequence-discriminative training techniques: maximum mutual information (MMI), minimum phone error (MPE), boosted-MMI (bMMI), and state-level minimum Bayes risk (sMBR). It has been demonstrated by selection of an optimal value of parameters using lattice generation, and adjustments of learning rates. In proposed framework, four different systems have been tested by analyzing various feature extraction approaches (with or without speaker normalization through Vocal Tract Length Normalization (VTLN) approach in test set) and classification strategy on with or without artificial extension of train dataset. To compare each system performance, true matched (adult train and test—S1, child train and test—S2) and mismatched (adult train and child test—S3, adult + child train and child test—S4) systems on large adult and very small Punjabi clean speech corpus have been demonstrated. Consequently, gender-based in-domain data augmented is used to moderate acoustic and phonetic variations throughout adult and children’s speech under mismatched conditions. The experiment result shows that an effective framework developed on PNCC + VTLN front-end approach using TDNN-sMBR-based model through parameter optimization technique yields a relative improvement (RI) of 40.18%, 47.51%, and 49.87% in matched, mismatched and gender-based in-domain augmented system under typical clean and noisy conditions, respectively.

Comparison of Text Independent Speaker Identification Systems using GMM and i-Vector Methods

Voice biometric provides a person’s unique identity and hence is widely employed for security applications. Here, Text Independent Speaker Recognition system is implemented using Gaussian Mixture Models (GMM) and i-Vector method with two features PNCC (Power Normalized Cepstral Coefficients) and RASTA PLP (Relative Spectral Perceptual Linear Prediction) coefficients. It was observed that accuracy for speaker identification is better when pitch and formants are appended to basic features. Also, the accuracy of i-vector method with PLDA (Probabilistic Linear Discriminant Analysis) classifier is better than that with CDS (Cosine Distance Scoring) classifier. Further, the performance improves when longer utterances are used.

Comparison of Several Acoustic Modeling Techniques for Speech Emotion Recognition

Automatic Speech Emotion Recognition (SER) is a current research topic in the field of Human Computer Interaction (HCI) with a wide range of applications. The purpose of speech emotion recognition system is to automatically classify speaker's utterances into different emotional states such as disgust, boredom, sadness, neutral, and happiness. The speech samples in this paper are from the Berlin emotional database. Mel Frequency cepstrum coefficients (MFCC), Linear prediction coefficients (LPC), linear prediction cepstrum coefficients (LPCC), Perceptual Linear Prediction (PLP) and Relative Spectral Perceptual Linear Prediction (Rasta-PLP) features are used to characterize the emotional utterances using a combination between Gaussian mixture models (GMM) and Support Vector Machines (SVM) based on the Kullback-Leibler Divergence Kernel. In this study, the effect of feature type and its dimension are comparatively investigated. The best results are obtained with 12-coefficient MFCC. Utilizing the proposed features a recognition rate of 84% has been achieved which is close to the performance of humans on this database.

Hybrid Techniques for Arabic Letter Recognition

In this paper we investigate the use of the feed-forward back propagation neural networks (FFBPNN) for automatic speech recognition of Arabic letters with their four vowels (Fatha, dhamma, Kasra, Soukoun). This investigation will constitute a basically step for the recognition of continuous Speech. Features were extracted from recorded corpus by using a variety of conventional methods such as Linear Predictive Codes (LPC), Perceptual Linear Prediction (PLP), Relative Spectral Perceptual Linear Prediction (RASTA-PLP), Mel Frequency Cepstral Coefficients (MFCC), Continuous Wavelet Transform (CWT), etc. Here, several hybrid methods have been used too. Since the extracted features have large dimensionalities they were reduced by conserving the most discriminatory information with the Principal Component Analysis (PCA) technique. The recognition performance has been improved particularly when we use the PLP method followed by PCA technique.

Comparison and combination of features in a hybrid HMM/MLP and a HMM/GMM speech recognition system

Recently, the advantages of the spectral parameters obtained by frequency filtering (FF) of the logarithmic filter-bank energies (logFBEs) have been reported. These parameters, which are frequency derivatives of the logFBEs, lie in the frequency domain, and have shown good recognition performance with respect to the conventional mel-frequency cepstral coefficients (MFCCs) for hidden Markov models (HMM) based systems. In this paper, the FF features are first compared with the MFCCs and the relative spectral perceptual linear prediction (Rasta-PLP) features using both a hybrid HMM/MLP and a usual HMM/Gaussian mixture models (HMM/GMM) based recognition system, for both clean and noisy speech. Taking advantage of the ability of the hybrid system to deal with correlated features, the inclusion of both the frequency second-derivatives and the raw logFBEs as additional features is proposed and tested. Moreover, the robustness of these features in noisy conditions is enhanced by combining the FF technique with the Rasta temporal filtering approach. Finally, a study of the FF features in the framework of multistream processing is presented. The best recognition results for both clean and noisy speech are obtained from the multistream combination of the J-Rasta-PLP features and the FF features.