Speaker Verification Based on Different Vector Quantization Techniques with Gaussian Mixture Models

This paper describes the implementation of a speaker identification system with reference to Assamese language. The database consists of speech samples that were collected from 15 (fifteen) speakers for ten Assamese words representing the Assamese digits from 0 (shounyo) to 9 (no). Mel Frequency Cepstral Coefficients (MFCC) are used as features for this study. Two independent speaker identification systems have been built in this paper using Vector Quantization (VQ) and I-vector technique. The system built using the I-vector technique obtains comparatively better identification accuracy for speakers when compared with the system developed using VQ technique. Three different systems have been built for both the techniques based on variable feature size. A maximum accuracy of 92.38% is achieved using I-vector technique with 39 MFCC features.

The author reviews major methods of applying the vector quantization (VQ) technique to speech and speaker recognition. These include speech recognition based on the combination of VQ and the DTW/HMM (dynamic time warping/hidden Markov model) technique. VQ-distortion-based recognition, learning VQ algorithms, speaker adaptation by VQ-codebook mapping, and VQ-distortion-based speaker recognition. It is concluded that not only has the VQ technique reduced the amount of computation and storage, but it has also created new ideas for solving various problems in speech/speaker recognition. >

Enhancements to DTW and VQ decision algorithms for speaker recognition

The generalized Lloyd algorithm (GLA), better known as the Linde-Buzo-Gray (LBG) algorithm, is the most widely used technique in classical vector quantization (VQ) for speech or image signal compression. However, the encoding complexity of the algorithm grows exponentially with the product of coding rate and vector dimension, which prohibits applying the technique to tasks with moderate to large encoding rates or vector dimensions. The authors present a new VQ scheme which overcomes the successive search coding computation of traditional techniques by using a quasi-parallel mapping technique and makes VQ practical for higher encoding rates and/or vector dimensions. Neural computing techniques are used to implement parallel encoding and decoding mappings in VQ and the developed algorithm was applied to quantizing Gauss-Markov processes and artificial data sources. Comparisons of performance with the LBG algorithm are given. >

Gaussian Mixture models is one of the most popular statistical methods in speaker recognition. The purpose of this research is to perform experiments of speaker recognition using various feature vectors: four formants, four formants with fundamental frequency and mel cepstrum coefficients. Gaussian mixture models using mel cepstrum coefficients is baseline in speaker recognition and gives one of the best results in text independent speaker recognition. After implementing experiments of speaker recognition and comparing experimental results we can affirm that mel scale cepstral coefficients and four formants with fundamental frequency gives quite the same recognition accuracy, but creating of Gaussian mixture speaker models and recognition process continues a few times longer using mel scale cepstral coefficients, because count of calculations is few times greater in that case. Using only four formants gives the worst results of recognition accuracy. Ill. 7, bibl. 12 (in English; summaries in English, Russian and Lithuanian).

A FAST ALGORITHM FOR VECTOR QUANTIZATION: APPLICATION TO CODEBOOK GENERATION IN IMAGE SUBBAND CODING

Some applications of neural network algorithms for the design of vector quantizer codebooks are presented. Vector quantization has been applied to the problem of displaying natural images with a reduced set of colors (color palette) and to the interframe coding of image sequences. Self-organizing feature maps and competitive learning (CL) algorithms have been used for the codebook design and their results are compared with those obtained using a classical LBG (Linde, Buzo, Gray) algorithm. The best results are obtained using a CL algorithm with a new initialization strategy. With this technique the codebook design is very fast, and some experiments have been carried out in order to introduce adaptive vector quantization in the interframe coder. Frame by frame the codebooks used to quantize the most important information are updated and then the more active parts of the motion-compensated luminance differences are quantized with these new codebooks. A considerable rise of the coder performance was obtained. >

In this artificial intelligence time, speaker recognition is the most useful biometric recognition technique. Security is a big issue that needs careful attention because of every activities have been becoming automated and internet based. For security purpose, unique features of authorized user are highly needed. Voice is one of the wonderful unique biometric features. So, developing speaker recognition based on scientific research is the most concerned issue. Nowadays, criminal activities are increasing day to day in different clever way. So, every country should have strengthen forensic investigation using such technologies. The study was done by inspiration of contextualizing this concept for our country. In this study, text-independent Amharic language speaker recognition model was developed using Mel-Frequency Cepstral Coefficients to extract features from preprocessed speech signals and Artificial Neural Network to model the feature vector obtained from the Mel-Frequency Cepstral Coefficients and to classify objects while testing. The researcher used 20 sampled speeches of 10 each speaker (total of 200 speech samples) for training and testing separately. By setting the number of hidden neurons to 15, 20, and 25, three different models have been developed and evaluated for accuracy. The fourth-generation high-level programming language and interactive environment MATLAB is used to conduct the overall study implementations. At the end, very promising findings have been obtained. The study achieved better performance than other related researches which used Vector Quantization and Gaussian Mixture Model modelling techniques. Implementable result could obtain for the future by increasing number of speakers and speech samples and including the four Amharic accents.

A new vector quantization technique, called storage- and entropy-constrained multi-stage vector quantization (SECMSVQ) technique, is presented for progressive image transmission (PIT) applications. In the SECMSVQ technique, the rate constraint at each stage of the vector quantizer (VQ), and the storage size of the whole VQ are specified independently before the design. A dynamic programming technique is then used to allocate the storage size to each stage of the VQ optimally so that the quantization distortion of the VQ is minimized. The SECMSVQ is well-suited to constructing a PIT system where the transmission time and storage size are the major concerns. Moreover, simulation results show that the SECMSVQ technique can be used effectively to implement signal compression systems under the storage and rate constraints.

Summary form only given. Vector quantization (VQ),is a well-known compression technique which has been widely used in many speech and image coding systems. Techniques for codebook design attempt to produce a codebook that is optimum for a given source. To date, the most widely used technique for VQ codebook design is the LBG (Linde-Buzo-Gray) algorithm. Madeim et al. (1999) show that an unsupervised neural network algorithm, referred to as SOA (self-organizing algorithm), provides good VQ codebooks, leading to reconstructed signals with better quality when compared to the ones obtained by using LBG codebooks. In this paper, an investigation is carried out to evaluate the inherent quality of SOA and LBG codebooks regarding the computational complexity of medium distortion encoding. The present work shows that the SOA codebooks overperforms the LBG codebooks in the sense that they yield a smaller average number of multiplications per sample for image VQ.

In the biometrics community, face and speaker recognition are mature fields in which several systems have been proposed over the past twenty years. While existing systems perform well under controlled recording conditions, mismatch caused by the use of different sensors or a lack of cooperation from the subject still significantly affects performance, especially in challenging scenarios such as in forensics. Furthermore, existing methods suffer from scalability issues, which prevents them from taking advantage of increasingly large amounts of training data. This is otherwise a promising approach to improve accuracy in such challenging scenarios. In this thesis we address these problems of mismatch and complexity by developing scalable probabilistic models that we apply to face, speaker and bimodal recognition. Our contributions are four-fold. First, we propose a unified framework for session variability modeling techniques based on Gaussian mixture models (GMM), that encompasses inter-session variability (ISV) modeling, joint factor analysis (JFA) and total variability (TV) modeling. Second, we propose a novel exact and scalable formulation of probabilistic linear discriminant analysis (PLDA), which is a probabilistic and generative framework that models between-class and within-class variations. This formulation solves a major scalability issue, by improving both the time complexity of the training procedure from cubic to linear with respect to the number of samples per class, and the complexity of the scoring procedure. Furthermore, the implementations of all the proposed techniques are integrated into a novel collaborative open source software library called Bob 1 that enforces fair evaluations and encourages reproducible research. Fourth and finally, large-scale experiments are conducted with all of the above machine learning algorithms on several databases such as FRGC for face recognition, NIST SRE12 for speaker recognition and MOBIO for bimodal recognition, showing competitive performances.

How to effectively convert a sequence of variable-length acoustic features to a fixed-dimension representation has always been a research focus in speaker recognition. In state-of-the-art speaker recognition systems, the conversion is implemented by concatenating the mean and the standard deviation of a sequence of frame-level features. However, a single mean and a single standard deviation are limited descriptive statistics for an acoustic sequence even with powerful feature extractors such as convolutional neural networks. In this paper, we propose a novel statistics pooling method that can produce more descriptive statistics through a mixture representation. Our approach is inspired by the expectation–maximization (EM) algorithm in Gaussian mixture models (GMMs). Instead of using traditional GMM style alignment, we novelly leverage modern deep learning tools to produce a more powerful mixture representation. The novelty includes: (1) unlike GMMs, the mixture assignments are determined by an attention network instead of the Euclidean distances between the frame-level features and explicit centers; (2) instead of using a single frame as input to the attention network, contextual frames are included to smooth out attention transition; and (3) soft-attention assignments are replaced by hard-attention assignments via the Gumbel-Softmax with straight-through estimators. With the proposed attention mechanism, we obtained a 13.7% relative improvement over vanilla mean and standard deviation pooling in the VOiCES19-eval set.

ABSTRACT: Speech is one of the most natural and efficient forms of communication among humans, and it is becoming increasingly integral to human-computer interaction in modern applications such as virtual assistants, smart devices, and automated customer service systems. Among the various technologies that leverage speech, speaker recognition stands out as a biometric method that identifies or verifies individuals based on the unique characteristics of their voice. This approach has gained significant traction due to its non-intrusive nature, ease of integration, and applicability across various domains such as security, forensics, and personalised user experiences. In this paper, we introduce a novel speaker recognition system specifically designed to identify speakers based on utterances in the Marathi language, a linguistically rich and widely spoken regional language in India. The system utilises Mel-Frequency Cepstral Coefficients (MFCCS) to extract distinguishing vocal features that closely mimic the human auditory perception. MFCCS are particularly effective in capturing speech’s phonetic and acoustic properties, making them a preferred choice in speech and speaker recognition tasks. Vector quantisation (VQ) is applied to optimise the feature set and reduce computational complexity. VQ compresses the high-dimensional MFCC feature vectors into representative clusters without significantly compromising accuracy, thereby enhancing system efficiency. These features are then processed using hidden Markov Models (HMMS), which are adept at modelling temporal sequences and dynamic variations in speech. HMMS offer a statistical framework that effectively captures the sequential nature of speech patterns, leading to more reliable speaker modelling and recognition.

In the study of speaker identification, Mel Frequency Cepstral Coefficient (MFCC) method is the best and most popular which is used to feature extraction. Further Vector Quantization (VQ) technique is used to minimize the amount of data to be handled and mapping vectors from a large vector space to a finite number of regions in that space in recent years. Voice, like other biometrics, cannot be forgotten or misplaced, unlike knowledge-based (e.g., password) or possession-based (e.g., key) access control methods. In the present work, modified Mel frequency cepstral coefficients using discrete fractional Fourier transform and vector quantization is obtained. The experimental results are analyzed with the help of MATLAB.

Relevance factor of maximum a posteriori adaptation for GMM–NAP–SVM in speaker and language recognition