Comparison of Several Acoustic Modeling Techniques for Speech Emotion Recognition

Khasi Speech Recognition using Hidden Markov Model with Different Spectral Features: A Comparison

The model-based speech enhancement method usually models types of noises in a prior and selects one noise model in the enhancement phase. In this paper, we study the modeling and selection (i.e. noise identification) algorithm for the model-base speech enhancement. For the noise signal features, Mel Frequency Cepstrum Coefficient (MFCC), Liner Prediction Coefficient (LPC), Linear Spectral Frequency (LSF) and Linear Predictive Cepstrum Coefficients (LPCC) are utilized. For the noise model, codebook, Gaussian Mixture Model (GMM) and Hidden Markov Model (HMM) are adopted. We couple these features and models and provide a comparative performance analysis of them. In the experiment, we use NOIZEUS database for noise modeling and testing. The experiment results show that MFCC and GMM is an excellent couple for noise identification, and LFS can take the place of LPC in the codebook-based speech enhancement.

Emotion is a natural feeling which is distinguished from reasoning or knowledge, it is a strong feeling derived from one’s circumstance or surroundings. With the increase in man to machine interaction, speech analysis has become an integral part in reducing the gap between physical and digital world. An important sub field within this domain is the recognition of emotion in speech signals, which was traditionally studied in linguistics and psychology. Speech emotion recognition is a field having diverse applications. When implemented the Speech Emotion Recognition (SER) will be able to understand different human emotion such as anger, fear, happiness, sadness etc. Speech is a medium of expression of one’s perspective or feelings to other. Emotion recognition from audio signal requires feature extraction and classifier training. The feature vector consists of elements of the audio signal which characterize speaker specific features such as tone, pitch, energy, which is crucial to train the classifier model to recognize a particular emotion accurately. Thus, with the help of SER we can make conversations between human and computer more realistic and natural. Automatic Speech Emotion Recognition (SER) is a current research topic in the field of Human Computer Interaction (HCI) with wide range of applications. The speech features such as, Mel Frequency cepstrum coefficients (MFCC) and Mel Energy Spectrum Dynamic Coefficients (MEDC) are extracted from speech utterance. The Support Vector Machine (SVM) is used as classifier to classify different emotional states such as anger, happiness, sadness, neutral, fear, from Berlin emotional database.

In this paper, based on the study of two speech recognition algorithms, two designs of speech recognition system are given to realize this isolated speech recognition mobile robot control system based on ARM9 processor. The speech recognition process includes pretreatment of speech signal, characteristic extrication, pattern matching and post-processing. Mel-Frequency cepstrum coefficients (MFCC) and linear prediction cepstrum coefficients (LPCC) are the two most common parameters. Through analysis and comparison the parameters, MFCC shows more noise immunity than LPCC, so MFCC is selected as the characteristic parameters. Both dynamic time warping (DTW) and hidden markov model (HMM) are commonly used algorithm. For the different characteristics of DTW and HMM recognition algorithm, two different programs were designed for mobile robot control system. The effect and speed of the two speech recognition system were analyzed and compared.

Comparison of Khasi speech representations with different spectral features and hidden Markov states

This paper presents a practical speaker recognition system based on multi-classifier structure. Multi-classifier structure overcomes the shortcomings of single classifier, such as low recognition rate, narrow application field and critical demand of environment. Additionally, multi-classifier provides a novel way of improvement of system performance. The involved classifiers include ANN (artificial neural networks), GMM (Gaussian mixed model), sub-band classifiers, etc. The input features of classifiers contain MFCC (mel frequency cepstrum coefficient), LPCC (linear prediction cepstrum coefficient). Multi-classifier confusion adopts CFM (classification figure of merit) principle as object function. In practical application, the recognition rate of the system achieves 94% in environment of super short wave (SNR 15db)

In this paper, traditional music time domain features and cepstrum domain features are extracted using the spectral center of mass, spectral energy, linear prediction cepstrum coefficients (LPCC) and Mel frequency cepstrum coefficients (MFCC). After that, the traditional music signal is normalized using a normalization algorithm based on the Short-Time Fourier Transform (STFT). Finally, the performance of music source separation is evaluated using NSDR. In this paper, the percentage of inheritance rate for the four parts of traditional vocal music, traditional instrumental music, and traditional drama music before digitization is less than or equal to 40%, 35.06%, and 31.25%, respectively. After digitization, their percentage of inheritance rate is greater than or equal to 86%, 93.51%, and 87.5%, respectively. The inheritance rate of the three kinds of traditional music after digitization increased in the interval of 56%. After digitization, the inheritance rate of three kinds of traditional music increased in the ranges of 56%-60%, 18.75%-31.25% and 56.25%-81.25%, respectively. This indicates that the inheritance rate of three types of traditional music increases dramatically after being processed by deep learning techniques. Obviously, the support of deep learning technology is indispensable to enhance the inheritance and innovation mechanisms of traditional music culture digitization.

A text-independent, closed-set speaker identification method is proposed in this paper. The method uses linear predictive cepstrum coefficients (LPCCs) as the measured features and follows general regression neural network (GRNN) approaches based on non-linear partition (NLP) algorithm and kernel principal component analysis (KPCA). The input speech signal is pre-emphasized, windowed, and LPC analyzed, resulting in a sequence of vectors of LPC derived cepstrum coefficients. To reduce the correlation and dimension of elements in the feature vector, the NLP algorithm is employed to partition the LPCCs into several segments. The dimensions of each LPCCs segment are reduced by KPCA, then fed to a GRNN for the classification of speaker identification. The numerical experiments are carried out to verify the theoretical results and clearly show that our identification system has good recognition ability in term of accuracy.

In this paper, we propose a new method for voice feature extraction by using a hierarchical clustering approach of inverse discrete cosine transform (IDCT) cepstrum coefficient. Since the IDCT cepstrum coefficient is transformed into the feature vector based on Euclidean metric, we call the proposed voice feature as “E-vector”. Comparing with other voice features, e.g. Mel frequency cepstrum coefficient (MFCC) and histogram of DCT cepstrum coefficients (HDCC), the E-vector can represent the dynamic characteristics and interactional relationships of the voice. In the numerical experiments, we use the speech files based on the voice of 630 people in TIMIT corpus, and employ Gaussian mixture model to compare the recognition accuracy of E-vector with that of MFCC and HDCC. The results show that E-vector outperformed other voice features on personal identification, and presented higher extensiveness on voice feature extraction.

In view of the growing use of automatic speech recognition in the modern society, we study various alternative representations of the speech signal that have the potential to contribute to the improvement of the recognition performance. In this paper wavelet based features using different wavelets are used for Hindi digits recognition. The recognition performance of these features has been compared with Linear Prediction Coefficients (LPC) and Perceptual Linear Prediction (PLP) features. All features have been tested using Hidden Markov Model (HMM) based classifier for speaker independent Hindi digits recognition. The recognition performance of PLP features is11.3% better than LPC features. The recognition performance with db10 features has shown a further improvement of 12.55% over PLP features. The recognition performance with db10 is best among all wavelet based features.

유도 전동기는 항공 산업, 자동차 산업 등의 산업 현장에서 중요한 역할을 하고 있으며, 이러한 유도 전동기의 고장으로 인한 피해를 최소화하기 위해 유도 전동기의 고장 검출 및 분류 시스템의 개발이 중요한 문제로 대두되고 있다. 이에 본 논문에서는 정상 및 각종 비정상 상태의 유도 전동기 진동 신호에 대해 부분 자기 상관(partial autocorrelation, PARCOR) 계수, 로그 스펙트럼 파워(log spectrum powers, LSP), 캡스트럼 계수의 평균값(cepstrum coefficients mean, CCM), 멜 주파수 캡스트럼 계수(mel-frequency cepstrum coefficient, MFCC)의 네 가지 특징 벡터를 신경 회로망의 입력으로 사용하여 유도 전동기의 고장을 검출하고 분류하였다. 고장 분류를 위한 최적의 특징 벡터를 찾기 위해 추출하는 특징의 수를 2에서 20으로 바꾸어 가며 분류 성능을 평가한 결과 CCM을 제외한 나머지의 경우 5~6의 특징만으로 분류 정확도가 거의 100%에 가까운 결과를 보였다. 또한 본 논문에서는 실제 산업 현장에서 진동 신호 취득 시 포함될 수 있는 잡음을 고려하여 취득한 신호에 백색 잡음(white Gaussian noise)을 인위적으로 추가하여 실험한 결과 LSP, PARCOR, MFCC 순으로 잡음 환경에 강인한 특징 벡터임을 확인할 수 있었다. Induction motors play a vital role in aeronautical and automotive industries so that many researchers have studied on developing a fault detection and classification system of an induction motor to minimize economical damage caused by its fault. With this reason, this paper extracts robust feature vectors from the normal/abnormal vibration signals of the induction motor in noise circumstance: partial autocorrelation (PARCOR) coefficient, log spectrum powers (LSP), cepstrum coefficients mean (CCM), and mel-frequency cepstrum coefficient (MFCC). Then, we classified different types of faults of the induction motor by using the extracted feature vectors as inputs of a neural network. To find optimal feature vectors, this paper evaluated classification performance with 2 to 20 different feature vectors. Experimental results showed that five to six features were good enough to give almost 100% classification accuracy except features by CCM. Furthermore, we considered that vibration signals could include noise components caused by surroundings. Thus, we added white Gaussian noise to original vibration signals, and then evaluated classification performance. The evaluation results yielded that LSP was the most robust in noise circumstance, then PARCOR and MFCC followed by LSP, respectively.

In recent years, crime prevention systems have been developed to detect various hazardous situations. In general, the systems utilize the image information recorded by a camera to monitor the situations. It is however difficult to detect them in the blind area. To address the problem, it is required to utilize not only image information but also acoustic information occurred in such situations. Our previous study showed that two acoustic features including rahmonic and mel-frequency cepstrum coefficients (MFCCs) are effective for detecting the shouted speech. Rahmonic shows a subharmonic of fundamental frequency in the cepstrum domain, and MFCCs represent coefficients that collectively make up mel-frequency cepstrum. In this method, a shouted speech model is constructed from these features by using a gaussian mixture model (GMM). However, the previous method with GMM has difficulty in representing temporal changes of the speech features. In this study, we further expand the previous method using hidden Markov model (HMM) which has state transition to represent the temporal changes. Through objective experiments, the proposed method using HMM could achieve higher detection performance of the shouted speech than the conventional method using GMM.

This paper compares the performance of two variants of the Particle Swarm Optimization (PSO) algorithm; PSO with constriction factor (PSO), and mutative PSO (MPSO) in optimizing Mel Frequency Cepstrum Coefficients (MFCC) parameters. The parameters were used to extract an optimal feature set for classifying healthy and hypothyroid infant cry using Multi-Layer Perceptrons (MLP). Specifically, the PSO variants optimize the number of filter banks and number of cepstrum coefficients in MFCC. Based on the values chosen by both PSO variants, the extracted features were then fed to a MLP classifier, which was trained to discriminate between the healthy and hypothyroid infant cry. Comparisons between the performance of PSO variants showed that MPSO managed to improve the convergence rate by 2.67% compared to PSO.

In the recent years, many research works have been published using speech related features for speech emotion recognition, however, recent studies show that there is a strong correlation between emotional states and glottal features. In this work, Mel-frequency cepstralcoefficients (MFCCs), linear predictive cepstral coefficients (LPCCs), perceptual linear predictive (PLP) features, gammatone filter outputs, timbral texture features, stationary wavelet transform based timbral texture features and relative wavelet packet energy and entropy features were extracted from the emotional speech (ES) signals and its glottal waveforms(GW). Particle swarm optimization based clustering (PSOC) and wrapper based particle swarm optimization (WPSO) were proposed to enhance the discerning ability of the features and to select the discriminating features respectively. Three different emotional speech databases were utilized to gauge the proposed method. Extreme learning machine (ELM) was employed to classify the different types of emotions. Different experiments were conducted and the results show that the proposed method significantly improves the speech emotion recognition performance compared to previous works published in the literature.

<p>In Ethiopia, the largest ethnic and linguistic groups are the Oromos, Amharas and Tigrayans. This paper presents the performance analysis of text-independent speaker identification system for the Amharic language in noisy environments. VQ (Vector Quantization), GMM (Gaussian Mixture Models), BPNN (Back propagation neural network), MFCC (Mel-frequency cepstrum coefficients), GFCC (Gammatone Frequency Cepstral Coefficients), and a hybrid approach had been use as techniques for identifying speakers of Amharic language in noisy environments. For the identification process, speech signals are collected from different speakers including both sexes; for our data set, a total of 90 speakers’ speech samples were collected, and each speech have 10 seconds duration from each individual. From these speakers, 59.2%, 70.9% and 84.7% accuracy are achieved when VQ, GMM and BPNN are used on the combined feature vector of MFCC and GFCC. </p>