Santosh Gaikwad

A Review on Speech Recognition Technique

The Speech is most prominent & primary mode of Communication among of human being. The communication among human computer interaction is called human computer interface. Speech has potential of being important mode of interaction with computer .This paper gives an overview of major technological perspective and appreciation of the fundamental progress of speech recognition and also gives overview technique developed in each stage of speech recognition. This paper helps in choosing the technique along with their relative merits & demerits. A comparative study of different technique is done as per stages. This paper is concludes with the decision on feature direction for developing technique in human computer interface system using Marathi Language.

Performance Evaluation of Speech Synthesis Techniques for English Language

The conversion of text to synthetic production of speech is known as text-to-speech synthesis (TTS). This can be achieved by the method of concatenative speech synthesis (CSS) and hidden Markov model techniques. Quality is the important paradigm for the artificial speech produced. The study involves the comparative analysis for quality of speech synthesis using hidden Markov model and unit selection approach. The quality of synthesized speech is evaluated with the two methods, i.e., subjective measurement using mean opinion score and objective measurement based on mean square score and peak signal-to-noise ratio (PSNR). Mel-frequency cepstral coefficient features are also extracted for synthesized speech. The experimental analysis shows that unit selection method results in better synthesized voice than hidden Markov model.

Feature extraction using fusion MFCC for continuous marathi speech recognition

This paper presents the performance of feature extraction techniques for speech recognition, for the classification of speech represented by a particular continuous sentence model. The goal of this study is to present independent as well as comparative performances of popular appearance based feature extraction techniques i.e. Linear Discriminative Analysed and Mel Frequency Cestrum Coefficient. Mel Frequency Cepstrum Coefficient (MFCC) helps us in extracting feature where as linear discriminant analysis (LDA) is used for reducing dimension of extracted feature. We experimented MFCC feature extraction individually and proposed a Fusion of MCCC and LDA for feature extraction.

Accent Recognition for Indian English using Acoustic Feature Approach

Accent is the basic pattern of acoustic feature and pronunciation. It can identify the person’s social and linguistic background. It is an important source of inter as well as intra speaker variability. The accent dependent dictionary or model can be used to improve accuracy of speech recognition system. In this study we present an experimental approach of acoustic speech feature for Marathi & Arabic accents for English speaking. The detail study of acoustics correlates the accent using formant frequency, energy and pitch characteristics. The database consists of speech from speaker with Marathi as their mother tongue and speakers from Iraq with Arabic language as mother tongue. Both the speakers were asked to speak English number from zero to nine. Through experimental results the fifth formant frequency found to be very effective for accent recognition.

Novel approach based feature extraction for Marathi continuous speech recognition

This paper examines the performance of feature extraction techniques and the classification of speech represented by continuous Marathi speech model. The goal of this study is to present independent as well as comparative performances of most popular appearance based feature extraction techniques i.e. Principal Component Analysis, Linear Discriminative Analysis, Mel Frequency Cepstrum Coefficient and Discrete Wavelet Transformation. To improve performance of speech recognition, it needed to perform dimensionality reduction of speech data. The Motivation was the lack of direct and detailed independent comparison in all possible algorithms. Mel Frequency Cepstrum Coefficient (MFCC) help us in extracting feature where as Principal Component Analysis (PCA), linear discriminate analysis (LDA), and Discrete Wavelet Transformation are used for reducing dimension of extracted feature. We experimented with several combinations by combining techniques such as MFCC+PCA (MFPCA), MFCC+LDA (MFLDA), MFCC+DWT (MFDWT), MFCC+PCA+DWT (MFPDWT) and MFCC+LDA+ DWT (MFLDWT).The performance of technique is compared on accuracy as well as real time factor. The MFCC+LDA+DWT (MFLDWT) is a best combination which provides the accuracy of 95.06% where real time factor is 0.35 second as compared to other combination.

Creation of Marathi speech corpus for automatic speech recognition

This paper describes the collection of audio corpus for Marathi language. Marathi is one of the regional Indian languages. There are 12 vowels and 36 consonants present in Marathi languages. The objective of the research is to create the speech corpus which can be used for automatic speech recognition system for various domains like telephonic inquiry system, teaching tutor etc. The size of corpus collected is 28420 isolated words and 17470 sentences from around 500 speakers. The speech utterances were recorded in 16 kHz in three recording medium, a headset, desktop mounted microphone and Mobile phone. The corpus is transcripted as well as annotated and is available for recognition system.

Design and Development of Marathi Speech Interface System

Speech is the most prominent and natural form of communication between humans. It has potential of being an important mode of interaction with computer. Man–machine interface has always been proven to be a challenging area in natural language processing and in speech recognition research. There are growing interests in developing machines that can accept speech as input. Normal person generally communicate with the computer through a mouse or keyboard. It requires training and hard work as well as knowledge about computer, which is a limitation at certain levels. Marathi is used as official language at government of Maharashtra. There is a need for developing systems that enable human–machine interaction in Indian regional languages. The objective of this research is to design and development of the Marathi speech Activated Talking Calculator (MSAC) as an interface system. The MSAC is speaker-dependent speech recognition system that is used to perform basic mathematical operation. It can recognize isolated spoken digit from 0 to 50 and basic operation like addition, subtraction, multiplication, start, stop, equal, and exit. Database is an essential requirement to design the speech recognition system. To reach up to the objectives set, a database having 22,320 sizes of vocabularies is developed. The MSAC system trained and tested using the Mel Frequency Cepstral Coefficients (MFCC), Linear Discriminative Analysis (LDA), Principal Component Analysis (PCA), Linear Predictive Codding (LPC), and Rasta-PLP individually. Training and testing of MSAC system are done with individually Mel Frequency Linear Discriminative Analysis (MFLDA), Mel Frequency Principal Component Analysis (MFPCA), Mel Frequency Discrete Wavelet Transformation (MFDWT), and Mel Frequency Linear Discrete Wavelet Transformation (MFLDWT) fusion feature extraction techniques. This experiment is proposed and tested the Wavelet Decomposed Cepstral Coefficient (WDCC) with 18, 36, and 54 coefficients approach. The performance of MSAC system is calculated on the basis of accuracy and real-time factor (RTF). From the experimental results, it is observed that the MFCC with 39 coefficients achieved higher accuracy than 13 and 26 variations. The MFLDWT is proven higher accuracy than MFLDA, MFPCA, MFDWT, and Mel Frequency Principal Discrete Wavelet Transformation (MFPDWT). From this research, we recommended that WDCC is robust and dynamic techniques than MFCC, LDA, PCA, and LPC. MSAC interface application is directly beneficial for society people for their day to day activity.