A SURVEY: SPEECH EMOTION IDENTIFICATION

A SURVEY: SPEECH EMOTION IDENTIFICATION Sejal Patel 1, Salman Bombaywala 2 M.E. Students, Department Of EC, SNPIT & RC, Umrakh, Gujarat, India 1 Assistant Professor, Department Of EC, SNPIT & RC, Umrakh, Gujarat, India 2 Abstract: In recent years a great deal of research has been done to automatically recognize emotions from human speech. Emotion is a symbol to express the feelings of one s expression. It is also important to identify the emotion of the user to understand the feeling of an utterance. There are main three goals in speech emotion detection. First is to select appropriate features like energy, the pitch, LPC, LPCC, voice quality, MFCC, and delta MFCC for emotional speech recognition. Second goal is to provide data-base which includes different languages, number of persons, and number of emotion. Third goal is to classifying speech into emotional states. It utilizes Hidden Markov Model (HMM), Artificial Neural Network (ANN), and Support Vector Machine (SVM), correlation, Dynamic Time Wrapping (DTW). The paper surveys different techniques developed by keeping the above goals into consideration. The emotions considered in this study are anger, fear, happy, neutral and sad. Speech emotion recognition is useful for applications which require man machine interaction such as web movies and computer tutorial where the response of those systems to the user depends on the detected emotion. It is also useful for in-car board system where information of the mental state of the driver may be provided to the system to initiate his/her safety. Key Words: ANN, correlation, DTW, End point, MFCC, Start point. I. INTRODUCTION Speech is communication or expression of thoughts in spoken words. Humans also express their emotion via written and spoken language. Emotion is a symbol to express the feelings of one s expression. We are still Far from having a natural interaction between man and machine because the machine does not understand the emotional state of the speaker. There are several methods of feature extraction like MFCC, energy, pitch, LPC (Linear Predictive Coding), voice quality etc., several classification techniques like ANN, HMM, SVM, correlation, DTW etc. which can be effectively utilized for analysis of voice signal. The acoustic variability introduced by the existence of different sentences, speakers, speaking styles, and speaking rates adds another obstacle because these properties directly affect most of the common extracted speech features such as frequency, pitch, and energy. First human voice is converted into digital signal form to produce digital data representing each level of signal at every discrete time step. After that digitized Speech samples are processed using combination of features like start point, end point to produce voice features. Then after these voice All rights reserved by www.ijaresm.net ISSN : 2394-1766 1

features can go through to select the pattern that matches the database and find correlation between each reference database and test input file in order to minimize the resulting error between them depending on feature. Speech emotion recognition is useful for application in car board system which gives the information of the mental state of the driver and provide to the system to initiate his /her safety. It is also useful for natural men-machine interaction such as computer tutorial application where the response of this system to the user depends on the detected emotion. It may be also useful in call center application and mobile communication [8]. It is useful in storytelling, interactive movies, to check the person s behavior, e-tutoring, call analysis in emergency services like ambulance and fire brigade. II. BLOCK DAIGRAM The basic block diagram of speech emotion detection is as shown in Figure 1. Figure 1 shows the overall process of speech recognition system [4]: Figure 1 Architecture of speech emotion detection [4] A. Input Speech: Speech contains information about textual messages speak identity and intended emotion. Speech is a complex signal produce from a time varying vocal tract system exited by a time varying excitation source. Speech is the fastest and efficient method of interaction between human. Speech is communication or expression of thoughts in spoken words. B. Preprocessing: In the pre-processing, work with sampling, frames, windows, start pointend point detection. The input signals are changed to suit the speaker recognition system by these works. We can Say that, it is a data compression processing, in that we are selecting the useful information from signal data available. The use of energy threshold is to remove the low energy frame which is usually the silence period of the speech signal. The silence frame will influence the value of features in the feature extraction process. We will compare the results when using different energy thresholds. 1. Energy of a signal The energy is a set of samples and it is related by the sum of the square of the samples. Find information about the time-dependent properties of the speech signal. It is very useful to select the threshold for start-point & end-point detection. To calculate STE the speech signal is sampled using a rectangular window function of width ω samples, where. Within each window, energy is computed as follow [6]: All rights reserved by www.ijaresm.net ISSN : 2394-1766 2

Where, e = Energy of a particular window xi = Indicating i th sample 2. Start and End Point Detection It gives a more beneficial as they are used to remove background noise and made speech Signal better than previous signal. Start point of any voice signal provide the exact starting location of voice sample based on STE values, so that all previous unwanted samples would be removed and new voice signal would be created. [6] This same process is also applied to detect End points of any speech signal. [9] C. Feature Extraction Technique: Feature Extraction Technique is a convert speech signal into a sequence of feature vector.the important part of feature extraction is to extract characteristics from the signal that are unique to each individual. There are different feature extraction techniques like Pitch, Energy, Vocal track cross section area, Speech rate, Formants, Mel Frequency Cepstral Coefficients (MFCC), Linear Prediction coefficient (LPC), Linear Prediction Cepstral Coefficients (LPCC), are used for emotion based speaker recognition system. 1. MFCC MFCC is the most popular feature extraction technique for speech recognition. (1) Figure 2 Block diagram of MFCC [3] Step 1: Pre-Emphasis: Pre-Emphasis is a technique used in speech processing to enhance high frequencies of the signal. It is done by using FIR high-pass filter.[4] 1. The speech signal generally contains more speaker specific information in high frequencies than the lower frequencies. 2. Pre-emphasis removes some of the glottal effects from the vocal tract parameter. Step 2: Framing: The continuous voice signal is divided into frames of N samples. Adjacent frames are being separated by M (M<N). [3] Generally the frame size is equal to power of two in order to facilitate the use of FFT. Successive frames are overlapping with each other by M samples. Overlapped frame is used for maintaining continuity between frames. Overlapping ensure high correlation between coefficient of successive frames. [6] Step 3: Windowing: Window is used as window shape by considering the next block in feature extraction processing chain and integrates all the closest frequency lines. [6] To All rights reserved by www.ijaresm.net ISSN : 2394-1766 3

window each individual frame means to minimize the signal discontinuities at the beginning and end of each frame. To minimize the spectral distortion by using the window to taper the signal to zero at the beginning and end of each frame. If window is defined as, W (n), 0 n N-1 (2) Where, N = number of samples in each frame W(n) = Hamming window Then the result of windowing signal is Y[n] =X[n] W [n] ; 0 n L 1 (3) Where, W (n) = 0.54 0.46 cos ; 0 n N-1 (4) = 0 herwise W [n] = Window operation Y[n] = Output signal X [n] = Input signal L = Number of samples per frame Step 4: Fast Fourier Transform (FFT): FFT is converting each frame of L samples from time domain into frequency domain. FFT reduces computation time required to compute a DFT and improve performance. The Fourier Transform is to convert the convolution of the glottal pulse U[n] and the vocal tract impulse response H[n] in the time domain. This statement supports the equation below: y(w) = FFT[h(t) x(t)] = H(w) X(w) (5) Where, H(w), x(w) and y(w) are the Fourier transform of h(t), x(t) and y(t) respectively. [6] Step 5: Mel Filter Bank: we can say that the frequencies range in FFT spectrum is very wide and voice signal does not follow the linear scale. Thus for each tone with an actual frequency f, Measured in Hz, a subjective pitch is measured on a scale called the mel scale [3]. Set of triangular filters that are used to compute a weighted sum of filter spectral components so that the output of process approximates to a Mel scale. Use the following approximate formula to compute the mel for a given frequency fin Hz: [3] (f) = 2595 log10(1 + f(hz) 700) (6) Figure 3 Mel scale filter bank [3][6] Above figure depicts a set of triangular filters that are used to compute a weighted sum of filter spectral components so that the output of process approximates to a Mel scale. Each filter s magnitude frequency is triangular in shape and equal to unity at the center frequency and decline linearly to zero at center frequency of two adjacent filters. Then, each filter output is the sum of its filtered spectral Components. All rights reserved by www.ijaresm.net ISSN : 2394-1766 4

Step 6: Log and Discrete Cosine Transform (DCT): Output of the mel filter bank is spectral components which is given to the Log block. This log Mel spectrum is converted into time domain using Discrete Cosine Transform (DCT). The result of the reconstruction is called Mel-Frequency Cepstrum Coefficient. The collection of coefficient is called acoustic vectors. So that, each input utterance is transformed into a sequence of acoustic vector. [6] 2. LPC: LPC Coefficients can closely approximate current speech sample as a linear combination of past samples, the LPC coefficients cannot totally represent the current speech sample. Because the human speech production system is not linear while the LPC model is, the LPC coefficients only represent the linear property and loss the nonlinear part. Therefore, the LPCC coefficients were proposed for represent the nonlinear property. D. Classifier: Different types of classifier are used, such as, Polynomial classifier, Hidden Markov Model (HMM), Gaussian Mixture Model (GMM),correlation, Artificial Neural Network (ANN), Support Vector Machine (SVM), Dynamic Time Wrapping (DTW) is available. Here, MFCC is used as basic feature extraction techniques with a classifier Artificial Neural Network (ANN). 1. Correlation: The most familiar measure of dependence between two quantities is the person product moment correlation coefficient, or person s correlation coefficient, commonly called simply the correlation coefficient. It is obtained by dividing the covariance of the two variables by the product of their standard deviations. Karl person developed the coefficient. Person correlation coefficient can be the correlation coefficient is symmetric: Corr (X,Y)= corr (Y,X) (7) The person correlation is +1 in the case of a perfect direct (increasing) linear relationship (correlation), -1 in the case of a perfect decreasing (inverse) linear relationship (anticorrelation), and some value between -1 and 1 in all other cases, indicating the degree of linear dependence between the variable. As it approaches zero there is less of a relationship. The closer the coefficient is to either 1 or 1, the stronger the correlation between the variables. [12] 2. ANN: Artificial Neural Network (ANN) is a system which operates in a very similar manner as that of the human brain, and solves problems by self learning.[1] NN is referred to a network of biological neurons system that process and transmit information. An ANN is configured for a specific application such as pattern recognition or data classification through learning process. Learning in biological system involves adjustments to the synaptic connections that exist between the neurons. Artificial neural network (ANN) possesses excellent discriminates power and learning capabilities and represents implicit knowledge. Therefore, ANN is used in emotion recognition fields. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurons. Artificial neural networks are the result of academic investigation that uses mathematical formulations to model nervous system operations. The resulting techniques are being successfully applied in a variety of everyday business applications. 3. Dynamic Time Warping (DTW): All rights reserved by www.ijaresm.net ISSN : 2394-1766 5

DTW algorithm is based on measuring similarity among two time series which may vary in time or speed. The similarity is evaluated in terms of alignment between two times series if one time series may be warped nonlinearly by stretching or shrinking it along its time axis [6][ 9]. This warping between two time series can then be used to find analogues regions between two time series or to determine similarity between the two time series. Mathematically, the DTW contrast two dynamic patterns and evaluates similarity by calculating a minimum distance between them. To realize this, consider two time series Q and C, which has length n and m respectively. Where, Q = q1, q2, qi,... qn C = c1, c2, cj,... cm To align two sequences using DTW, an n -by- m matrix where the (i th, j th ) element of the matrix involves the distance d (qi, cj) among the two points qi and cj is established. Then, the absolute distance between the values of two sequences is calculated using the Euclidean distance computation [4]: d(qi,cj)= 2 (qi - cj) (9) Each matrix element (i, j) corresponds to the alignment between the points qi and cj. Then, accumulated distance is obtained by: D (i, j) = min (D (i -1, j -1), D (i -1, j), D (i, j -1)] + D (i, j) This is done as follows [6]: 1. Start with the calculation of g (1, 1) = d (1, 1). Calculate the first row g (i, 1) =g (i 1, 1) + d(i, 1). 2. Calculate the first column g (1, j) = g (1, j) + d (1, j). 3. Move to the second row g (i, 2) = min (g (i, 1), g (I 1, 1), g (i 1, 2)) + d (i, 2). Book keep for each cell the index of this neighboring cell, which contributes the minimum score. 4. Carry on from left to right and from bottom to top with the rest of the grid g (i, j) = min (g (i, j 1), g (i 1, j 1), g (i 1, j)) + d (i, j). 5. Trace back the best path through the grid starting from g (n, m) & moving towards g (1,1). Hence the path which gives minimum distance after testing with the feature vectors stored in the database is the identified speaker emotions. COMPARISON TABLE OF SURVEY PAPER Sr.No. Features Classifier Remarks 1 MFCC, DWT, LPC, f0, voice energy ANN ANN is better but it is language dependant 2 MFCC, formants BP neural Network, Low accuracy 3 MFCC - MFCC gives better result MFCC, pitch, GMM gives the better performance for 4 GMM, HMM anger and surprise, HMM is for disgust, energy fear and neutral 5 MFCC,STE ZCR Correlation, DTW DTW & correlation are finding the best match proves to be an effective method for speech recognition All rights reserved by www.ijaresm.net ISSN : 2394-1766 6

III. CONCLUDING REMARKS Speech emotion recognition systems based on the several classifiers is illustrated. The important issues in speech emotion recognition system are the signal processing unit in which appropriate features are extracted from available speech signal and another is a classifier which recognizes emotions from the speech signal. Different types of feature and classification techniques are studied. By the analysis it can be said that MFCC is the best feature extraction technique for emotion recognition. There are many applications like: To check the persons behavior, E-tutoring, Call analyses in emergency services like ambulance and fire brigade, Emotion Recognition in Call Center, Storytelling, In-Car Board System, Computer Games, and In Robots. REFERENCES [01] Firoz Shah.A, Raji Sukumar.A, Babu Anto.P, Automatic Emotion Recognition from Speech Using Artificial Neural Networks with Gender- Dependent Databases, International Conference on Asdvances in Computing, Control, and Telecommunication Technologies, IEEE, 2009 [02] Ying SHI And Weihua SONG, Speech Emotion Recognition Based on Data Mining Technology, International Conference on Natural Computation (ICNC 2010), IEEE, 2010 [03] Anurag Jain, Nupur Prakash, Evaluation of MFCC for Emotion Identification in Hindi Speech, IEEE, 2011 [04] Tsang-Long Pao, Chun-Hsiang Wang, and Yu-Ji Li, A Study on the Search of the Most Discriminative Speech Features in the Speaker Dependent Speech Emotion Recognition, International Symposium on Parallel Architectures, Algorithms and Programming, IEEE, 2012 [05] Manav Bhaykar, Jainath Yadav, and K. Sreenivasa Rao, Speaker Dependent, Speaker Independent and Cross Language Emotion Recognition from Speech Using GMM and HMM, IEEE, 2013 [06] Nidhi Desai, Kinnal Dhameliya and Vijayendra Desai, Recognizing voice commands for robot using MFCC and DTW, International Journal of Advanced Research in Computer and Communication Engineering, May 2014 [07] Vinay, Shilpi Gupta, Anu Mehra, Gender Specific Emotion Recognition Through Speech Signals, International Conference on Signal Processing and Integrated Networks (SPIN), 2014 [08] Tatjana Liogien, Gintautas Tamuleviius, Minimal Cross-correlation Criterion for Speech Emotion Multi-level Feature Selection, IEEE, 2015 [09] S.Lalitha, Anoop Mudupu, Bala Visali Nandyala, Renuka Munagala, Speech Emotion Recognition using DWT, IEEE, 2015 [10] Ritu D.Shah, Dr. Anil.C.Suthar, Speech Emotion Recognition Based on SVM Using MATLAB, International Journal of Innovative Research in Computer and Communication Engineering, 2016 [11] Lawrence R. Rabiner and Ronald W. Schafer, Digital Processing of Speech Signals; 9th Edn; Dorling Kindersley Pvt. Ltd., licepearson Education in South Asia, 2012 [12] https://statistics.laerd.com/statistical-guides/pearson-correlation-coefficient-statisticalguide.php (accessed on: 24 march 2017) All rights reserved by www.ijaresm.net ISSN : 2394-1766 7