Puja Ramesh Chaudhari ¹, John Sahaya Rani Alex ¹

Affiliations
1 School of Electronics Engineering, VIT University, Chennai - 632014, Tamil Nadu, IN

Abstract

Background/Objective: Speech is one of the modes for Human Computer Interface (HCI). Speech contains message to convey as well as the speaker characteristics such as speaker identity and emotional state of the speaker. Recently, researchers are taking more interest in the emotional parameters of speech signals which helps to improve the functionality of HCI. This research focus on selecting features which helps to identify the emotion of the speaker. Methods/Statistical Analysis: Mel Frequency Cepstrum Coefficient (MFCC), Linear Prediction Cepstrum Coefficient (LPCC) and Perceptual Linear Predictive (PLP) methods are used to extract the features. Each emotion is modeled as one Hidden Markov Model (HMM) using Hidden Markov Tool Kit (HTK tool kit). The Beagle Bone Black (BBB) board is chosen for the implementation because of the form factor. Findings: The results indicate that MFCC features gives 100% accuracy for surprise emotion, PLP features gives 100% accuracy for anger emotion and LPCC features give 100% accuracy for fear emotion. Conclusion/Improvement: A hybrid feature extraction method should be devised to detect all emotions with 100% accuracy.

Keywords

BBB, Emotion recognition, HCI, HMM, LPCC, MFCC.

Full Text

   

John Sahaya Rani Alex ¹, Ajinkya Sunil Mukhedkar ¹, Nithya Venkatesan ¹

Affiliations
1 School of Electronics Engineering Department, VIT University, Chennai - 600 127, Tamil Nadu, IN

Abstract

Background: Speech recognition is an active area of research, used to transliterate words vocalized by individuals in order to make them machine recognizable. For real time speech recognition applications the response time, size of training data and recognition accuracy are the important aspects. Methods: A Hybrid speech recognition system is proposed on the basis on Artificial Neural Network (ANN) in this research. The Self Organising Feature Map (SOFM) is used to reduce the feature vector dimensions which are extracted using the Mel-Frequency Cepstrum Coefficients (MFCC), Perceptual Linear Predictive (PLP) and Discrete Wavelet Transform (DWT) methods. The Back Propagation Network (BPN) algorithm is used for training the Artificial Neural Network for pattern classification. Findings: The proposed method is tested with TIDIGITS data. Results indicate that despite of the large reduction in the feature vector dimensions the recognition accuracy obtained using SOFM technique is same as that of the recognition accuracy of the conventional methods. The response time is also fast and the data size of the input data is reduced considerably. The proposed hybrid system is further tested using multilingual isolated digit data.

Keywords

Artificial Neural Network, Discrete Wavelet Transform, Feature Extraction, Mel Frequency Cepstrum Co-efficients, Perceptual Linear Predictive, Self-organising Feature Map, Speech Recognition

Full Text

   

Tanay Mayankbhai Modi ¹, Patel Hardik Anilkumar ¹, John Sahaya Rani Alex ¹

Affiliations
1 School of Electronics Engineering Department, VIT University, Chennai Campus, Chennai - 600127, Tamil Nadu, IN

Abstract

Background/Objectives: Discrete wavelet Transform (DWT) is substantially applied in many Digital Signal Processing (DSP) applications. Multiplication of the coefficient is the key factor for complexity in FIR filters. Methods/Statistical Analysis: This paper presents different multiplication techniques used to reduce the complexity in DWT such as Constant Shift Method (CSM), Vedic multiplication and Binary Signed Sub-coefficient (BSS) method. CSM technique uses Shift and Add unit followed with Mux to select appropriate output. BSS technique uses signed sub-coefficients, which reduces the multiplexer size. Vedic multiplication is famous for its low complexity architecture of bit-bit multiplication. Partial product and BCSE methods are used for designing the architectures. The coefficients are represented in IEEE 754 floating point half precision standard. Software simulation is performed in ModelSim and the designs are synthesized in Cadence RC tool for 180 nm technology. Findings: The CSM method gives high speed operation because of its reduced number of adders. The Vedic method provides low power and high speed operation. The BSS method uses Multiplexers and signed bit as control switch, which has a great impact over area requirement. The area requirement is reduced by 6% in BSS technique. Vedic multiplier gives about 23% power reduction compared to conventional multiplier. Both CSM and BSS techniques can be called Reconfigurable architectures as they can be hardwired and the mux output depends on the coefficients only. Conclusion/Improvements: CSM technique is gives the perfect balance for power reduction and area efficiency. Vedic technique gives immense reduction in power consumption but the area overhead is increased.

Keywords

Binary Signed Sub-Coefficients, Constant Shift Method, Daubechies (db4) DWT, Discrete Wavelet Transform (DWT), Vedic Multiplication.

Full Text