[Paper Review] Voice Recognition Algorithms using Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time Warping (DTW) Techniques

Lindasalwa Muda, M. Humrosia Begam+1 more|arXiv (Cornell University)|Mar 22, 2010

Music and Audio ProcessingComputer Science6 references821 citations

TL;DR

This paper proposes a voice recognition system using Mel Frequency Cepstral Coefficients (MFCC) for feature extraction and Dynamic Time Warping (DTW) for sequence matching. It demonstrates that MFCC effectively captures perceptually relevant speech features while DTW compensates for temporal variations in speech, achieving accurate speaker recognition through non-linear alignment of speech patterns.

ABSTRACT

Digital processing of speech signal and voice recognition algorithm is very important for fast and accurate automatic voice recognition technology. The voice is a signal of infinite information. A direct analysis and synthesizing the complex voice signal is due to too much information contained in the signal. Therefore the digital signal processes such as Feature Extraction and Feature Matching are introduced to represent the voice signal. Several methods such as Liner Predictive Predictive Coding (LPC), Hidden Markov Model (HMM), Artificial Neural Network (ANN) and etc are evaluated with a view to identify a straight forward and effective method for voice signal. The extraction and matching process is implemented right after the Pre Processing or filtering signal is performed. The non-parametric method for modelling the human auditory perception system, Mel Frequency Cepstral Coefficients (MFCCs) are utilize as extraction techniques. The non linear sequence alignment known as Dynamic Time Warping (DTW) introduced by Sakoe Chiba has been used as features matching techniques. Since it's obvious that the voice signal tends to have different temporal rate, the alignment is important to produce the better performance.This paper present the viability of MFCC to extract features and DTW to compare the test patterns.

Motivation & Objective

• To develop a reliable and efficient voice recognition system using digital signal processing techniques.
• To address the challenge of temporal variability in speech signals, which complicates direct comparison of voice patterns.
• To evaluate the effectiveness of MFCC as a feature extraction method that models human auditory perception.
• To investigate DTW as a robust matching technique for non-linearly aligned speech sequences.
• To demonstrate the feasibility of combining MFCC and DTW for accurate automatic voice recognition.

Proposed method

• Preprocessing the raw speech signal to remove noise and enhance clarity before feature extraction.
• Applying Mel Frequency Cepstral Coefficients (MFCC) to extract spectral features that reflect human auditory perception.
• Using the discrete Fourier transform (DFT) and mel-scale filter banks to convert frequency spectra into perceptually weighted coefficients.
• Applying Dynamic Time Warping (DTW) to align and compare speech sequences with variable speaking rates.
• Implementing DTW with a cost function that minimizes the cumulative distance between feature vectors of test and reference signals.
• Matching test speech patterns against a database of stored reference templates using DTW to identify the closest match.

Experimental results

Research questions

• RQ1 Can MFCC effectively extract discriminative features from speech signals for voice recognition?
• RQ2 How well does DTW handle temporal variations in speech signals during pattern matching?
• RQ3 What is the performance of the MFCC-DTW combination in identifying speakers under varying speaking rates?
• RQ4 Is the MFCC-DTW approach more robust than other methods like LPC or HMM for isolated word recognition?
• RQ5 Can this method achieve high accuracy with minimal computational overhead in real-time applications?

Key findings

• MFCC provides a compact and perceptually relevant representation of speech signals by emphasizing critical auditory bands.
• DTW successfully aligns speech sequences with variable durations, improving matching accuracy despite differences in speaking speed.
• The combination of MFCC and DTW yields high recognition accuracy for isolated word or speaker recognition tasks.
• The method is computationally efficient and suitable for real-time applications due to its non-parametric nature.
• The system demonstrates robustness to variations in pitch and speaking rate, making it suitable for practical deployment.
• The approach outperforms traditional methods like LPC in terms of simplicity and recognition stability under variable conditions.