We present a novel approach to vector quantization in which a feature vector is represented by a binary vector. It is called binary quantization (BQ). The performance criterion of vector quantization, distortion (distance) measure, was employed for investigating the effectiveness of BQ. At 12 b/analysis frame, the average distortion caused by BQ is even lower than the intraspeaker average distance between two repetitions of the same word (after DTW alignment). Since the output of BQ is a binary sequence, it is possible to combine it with a forward Hamming net classifier. In terms of the idea of a hierarchical model for describing a speaker individual characteristics, a text-independent speaker identification system was set up. Experimental results show that the performance of this system is very good. Not only are the small memory space and little computation required, in the speaker identification system, but, more importantly, it shows strong robustness in additive Gaussian white noise.

Binary quantization of feature vectors for robust text-independent speaker identification

This paper presents three novel methods for speaker identification of which two methods utilize both the continuous density hidden Markov model (HMM) and the generalized fuzzy model (GFM), which has the advantages of both Mamdani and Takagi-Sugeno models. In the first method, the HMM is utilized for the extraction of shape-based batch feature vector that is fitted with the GFM to identify the speaker. On the other hand, the second method makes use of the Gaussian mixture model (GMM) and the GFM for the identification of speakers. Finally, the third method has been inspired by the way humans cash in on the mutual acquaintances while identifying a speaker. To see the validity of the proposed models [HMM-GFM, GMM-GFM, and HMM-GFM (fusion)] in a real-life scenario, they are tested on VoxForge speech corpus and on the subset of the 2003 National Institute of Standards and Technology evaluation data set. These models are also evaluated on the corrupted VoxForge speech corpus by mixing with different types of noisy signals at different values of signal-to-noise ratios, and their performance is found superior to that of the well-known models.

GFM-Based Methods for Speaker Identification

A novel two-level method is proposed for rapidly and accurately computing the line spectrum pair (LSP) frequencies. An efficient decimation-in-degree (DID) algorithm is also proposed in the first level, which can transform any symmetric or antisymmetric polynomial with real coefficients into the other polynomials with lower degrees and without any transcendental functions. The DID algorithm not only can avoid prior storage or large calculation of transcendental functions but can also be easily applied toward those fast root-finding methods. In the second level, if the transformed polynomial is of degree 4 or less, employing closed-form formulas is the fastest procedure of quite high accuracy. If it is of a higher degree, a modified Newton-Raphson method with cubic convergence is applied. Additionally, the process of the modified Newton-Raphson method can be accelerated by adopting a deflation scheme along with Descartes rule of signs and the interlacing property of LSP frequencies for selecting the better initial values. Besides this, Horner's method is extended to efficiently calculate the values of a polynomial and its first and second derivatives. A few conventional numerical methods are also implemented to make a comparison with the two-level method. Experimental results indicate that the two-level method is the fastest one. Furthermore, this method is more advantageous under the requirement of a high level of accuracy.

A novel two-level method for the computation of the LSP frequencies using a decimation-in-degree algorithm

A novel approach to establish membership functions of fuzzy states based on a functional-link neural network (FLNN) for text-independent speaker identification is proposed. Parameters in corresponding states derived from training utterances uttered by each speaker are fed to a fuzzy statistical model to form the histograms of the states. Data in normalized histograms are used as learning samples for the FLNNs. A set of FLNNs is used as a speaker's model. For each unknown speaker's voice in the test, parameters are fed to every FLNN corresponding to each fuzzy state respectively, the FLNN generalizes the degree of membership to the state, then the system scores the degree of membership to the speaker's model and chooses the speaker whose model's degree of membership is the highest.

A kind of fuzzy-neural networks for text-independent speaker identification

A new technique for text-independent speaker identification for noisy speech is presented. This technique is based on finding the ratio of the singular values of the feature vectors of the unknown speaker and each of the N reference features stored in the constructed database. The i/sup th/ reference feature that gives the largest ratio is considered as the feature of the unknown speaker. An overall correct recognition accuracy of 99.5% for clean speech and 77.5% for noisy speech of 0 dB SNR was obtained. It was found that, for clean and noisy speech of 15 dB SNR or less, the proposed technique outperforms the conventional matching measure such as the Euclidean and the weighted distances, respectively.

http://www.researchgate.net/profile/Rabah_Aldhaheri/publication/4072502_Text-independent_speaker_identification_in_noisy_environment_using_singular_value_decomposition/links/551c00d60cf2909047b999e8.pdf

Text-independent speaker identification in noisy environment using singular value decomposition

The difference in effectiveness in representing the individual features of speakers of line spectrum pair (LSP) frequencies and cepstrum derived from linear prediction analysis is demonstrated. A fuzzy mathematical algorithm is introduced. A fuzzy statistical method is used to build up the membership function as the speaker's template, and the maximum value of the membership function is used as the deciding criterion. A random combination of isolated digits from 0 to 9 is specified as the identification utterances. The system has been evaluated on a database of isolated digit utterances of 42 speakers, 20 males and 22 females, all university students. For each speaker, the training utterance time is approximately 36 seconds. The correct identification rate is approximately 99.7% for a random combination of five isolated digits. >

Y. Fang

Papers

Text independent speaker identification using fuzzy mathematical algorithm