Source-linked AI summary
Voice Recognition Algorithms using Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time Warping (DTW) Techniques
Lindasalwa Muda, Mumtaj Begam, I. Elamvazuthi
TL;DR
Voice recognition requires representing complex speech signals because direct analysis contains too much information. This paper uses MFCC to extract acoustic features and DTW to compare patterns, reporting optimal matching between test voices and stored reference templates and effective speaker authentication.
Problem
Complex speech signals contain too much information for direct analysis, motivating feature extraction and matching for automatic voice recognition.
Method
The paper extracts acoustic vectors with MFCC and uses DTW to nonlinearly align and compare feature-vector sequences.
Results
The test input voice matched optimally with the stored reference template through pairwise comparison of feature vectors and an optimal warping path.
Takeaways & Limitations
MFCC and DTW could be used effectively to authenticate a particular speaker from information contained in the voice signal.
Takeaways & Limitations
Training and testing signals can differ because of voice changes, health, speaking rate, noise, and recording-environment variation.
Abstract
from arXiv · showhide
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.
1 INTRODUCTION
Voice recognition converts complex speech waveforms into features that support identifying the speaker from an utterance. The paper focuses on MFCC for feature extraction and DTW for pattern comparison.
- Voice signal identification converts speech waveforms into coefficient-based features for further processing and cepstrum analysis.The feature representation captures time, frequency, and energy information.
- Voice recognition aims to determine which speaker is present from the individual’s utterance.
- MFCC and DTW are presented as techniques for comparing voice patterns and are implemented using MATLAB.The paper reports findings from a voice recognition study using these techniques.
2 PRINCIPLE OF VOICE RECOGNITION
The recognition system processes speech through preprocessing, MFCC feature extraction, and DTW-based matching. MFCC represents acoustic information using Mel-scaled cepstral coefficients, while DTW aligns sequences that differ in temporal rate.
- Voice Recognition Algorithms: The system has training and testing phases, with preprocessing, feature extraction, and recognition performed on microphone input.Operations include pre-emphasis, framing, windowing, Mel cepstrum analysis, and matching.
- Feature Extraction (MFCC): MFCC models human auditory perception using filter spacing that is linear below 1000 Hz and logarithmic above 1000 Hz.A Mel-frequency scale captures phonetic characteristics of speech.
- Feature Extraction (MFCC): The DCT converts the log Mel spectrum into acoustic vectors, and delta and double-delta features represent changes across frames.The method adds 13 delta features and 39 double-delta features associated with cepstral and energy changes.
- Feature Matching (DTW): DTW measures similarity between time series by nonlinearly stretching or shrinking one sequence to find an optimal alignment.The alignment uses an n-by-m distance matrix and an accumulated-distance path whose minimum determines the comparison distance.
- Feature Matching (DTW): DTW constrains candidate paths to monotonic, nondecreasing movements through the alignment grid and searches for a minimum-distance path.The sequence length is N and the number of templates is V in the stated polynomial-time formulation.
3 METHODOLOGY
The voice-recognition system accounts for recording and speaker variability, using a flowchart-based process for training and testing.
- Voice signals can differ across training and testing because of speaker changes, speaking rate, health, noise, and recording environment.
- Training requirements are documented as a process description in Table 1.
- The overall voice-recognition process is represented by a flowchart.
4 RESULT AND DISCUSSION
The evaluation uses voice signals and MFCC representations from different speakers, then applies DTW to compensate for temporal differences and match test inputs with stored templates.
- Voice analysis evaluates MFCC outputs from two different speakers.
- MFCC sequence lengths vary because constant window spacing and differing phoneme durations produce unequal, non-linear word timings.
- DTW compensates for sequence-length and timing differences by finding an optimal path with minimum accumulated distance between feature vectors.
- The test voice matched optimally with the reference template stored in the database in Figures 9–11.
- The matching result is consistent with pairwise comparison of feature vectors between incoming voice and template.
5 CONCLUSION
The paper concludes that MFCC and DTW can be used effectively for voice recognition and speaker authentication based on information in the voice signal.
- MFCC and DTW were effective for voice recognition and authenticating a particular speaker from individual voice information.
- LPC, HMM, and ANN are identified as techniques for investigation in future publications.