Speaker Identification Performance Research Articles

Improving Speaker Identification Performance Under the Shouted Talking Condition Using the Second-Order Hidden Markov Models

Speaker identification systems perform well under the neutral talking condition; however, they suffer sharp degradation under the shouted talking condition. In this paper, the second-order hidden Markov models (HMM2s) have been used to improve the recognition performance of isolated-word text-dependent speaker identification systems under the shouted talking condition. Our results show that HMM2s significantly improve the speaker identification performance compared to the first-order hidden Markov models (HMM1s). The average speaker identification performance under the shouted talking condition based on HMM1s is. On the other hand, the average speaker identification performance based on HMM2s is.

A discriminative training approach for text-independent speaker recognition

Gaussian mixture model (GMM) has been commonly used for text-independent speaker recognition. The estimation of model parameters is generally performed based on the maximum likelihood (ML) criterion. However, this criterion only utilizes the labeled utterances for each speaker model and very likely leads to a local optimization solution. To solve this problem, this paper proposes a discriminative training approach based on the maximum model distance (MMD) criterion. We investigate the characteristics of speaker recognition and further propose a novel selection strategy of competing speakers associated with it. Experimental results based on the KING and TIMIT databases demonstrate that our training approach was quite efficient to improve the performance of speaker identification and verification. When there were three training sentences for each speaker, the verification equal error rate (EER) of 168 speakers in TIMIT could be reduced by 30.4% compared with the conventional method.

Enhancement of GMM speaker identification performance using complementary feature sets

This paper describes a way of enhancing speaker identification (SiD) performance using N-best list method which utilises complementary feature sets. The SiD process is first done by training the Gaussian mixture model (GMM) classifier using paramcterised feature sets (PFS) to form speaker models. During testing, the likelihood of a speaker, given a set of speaker models is her score. Performance scores of SiD system is normally degraded as the population of speakers increases. This paper addresses this problem by using linear prediction cepstral coefficients (LPCC) to complement the results obtained from the PFS and the final identification is performed on a smaller population set. Results obtained using 2-best list indicate performance improvement.

Enhancing the Performance of Gaussian Mixture Model-Based Text Independent Speaker Identification

In this paper, we seek to enhance the identification performance of Gaussian Mixture Model (GMM)-based speaker identification systems in the presence of a limited amount of training data and a relatively large number of speakers. The performance is characterized by the identification accuracy, the identification time, and the model complexity. A new model order selection technique based on the Goodness of Fit (GOF) statistical test is proposed in order to increase the identification accuracy. This technique has shown to outperform other well known model order selection techniques like the Minimum Description Length (MDL) and the Akaike Information Criterion (AIC) in terms of the identification accuracy and the robustness against telephone channel degradation effects. In addition, the identification time is decreased by adapting the Linear Discriminative Analysis (LDA) feature extraction technique to fit our basic assumption of asymmetric multimodal distribution of the training data of each speaker. This modification results in a large decrease in the identification time with a little effect on the identification accuracy.

Enhancing the Performance of Gaussian Mixture Model-Based Text Independent Speaker Identification

In this paper, we seek to enhance the identification performance of Gaussian Mixture Model (GMM)-based speaker identification systems in the presence of a limited amount of training data and a relatively large number of speakers. The performance is characterized by the identification accuracy, the identification time, and the model complexity. A new model order selection technique based on the Goodness of Fit (GOF) statistical test is proposed in order to increase the identification accuracy. This technique has shown to outperform other well known model order selection techniques like the Minimum Description Length (MDL) and the Akaike Information Criterion (AIC) in terms of the identification accuracy and the robustness against telephone channel degradation effects. In addition, the identification time is decreased by adapting the Linear Discriminative Analysis (LDA) feature extraction technique to fit our basic assumption of asymmetric multimodal distribution of the training data of each speaker. This modification results in a large decrease in the identification time with a little effect on the identification accuracy.

Small group speaker identification with common password phrases

Text-dependent speaker identification performance is investigated for small groups of speakers in which each speaker in a group is assigned the same sentence-long password utterance. Password utterances are modeled by whole-phrase hidden Markov models (HMMs). Several model construction conditions are studied. Baseline maximum likelihood estimate (MLE) models are constructed from three same-session training utterances. Minimum classification error (MCE) models are constructed using the training utterances of all speakers in a group. In addition, models are constructed using additional test utterances from speakers in the group or additional utterances from speakers outside the group. Results show that error rates approximately double from 5-speaker groups to 10-speaker groups. MCE models provide about 25% improvement in closed- and open-set identification error rates, but less improvement, about 10% in imposter accept rates. The greatest improvements are obtained, for both MLE and MCE models, when customer test utterances augment the training utterances. For MCE models closed-set identification error rates are approximately 0.4% and 0.6% for 5- and 10-speaker groups, respectively, while imposter accept rates are approximately 4% and 10%, respectively, when customer reject rates are 5%.

Effectiveness of different vowel sounds in automatic speaker identification

Abstract: An experiment is performed to study the effectiveness of eleven English vowels (/ɜ/, /ʌ/, /u/, /ʊ/, /o/, /ɔ/, /a/, /ae/, /e/, / i /, /i/) in an automatic speaker identification task. The first four formant frequencies of the vowel sounds are used as parameters and a minimum distance classifier is used for speaker classification. The vowel /ɜ/ is found to give the best speaker identification performance.

Speaker recognition in conjunction with speaker independent word recognition

Our method for accomplishing speaker independent word recognition is to select a small set of word templates that typify and span individual speaker reference templates obtained from a large population of speakers. An unknown utterance is processed and compared with a set of such templates for each word in the vocabulary. Word recognition decision functions are based on combinations of template distances obtained in the comparison. In this study it is hypothesized that distributions of template distance scores are reasonably consistent for individual speakers and vary characteristically from speaker to speaker. This property is exploited to provide a speaker recognition capability in conjunction with speaker independent word recognition. It is shown that good speaker identification performance is dependent on the input of a sequence of distinct words. For a 20 speaker experimental population a mean speaker identification error less than 1% is obtained over a sequence of seven distinct words. Techniques are also described for grouping individual speaker references into class references.

Effects of selected vocal disguises upon spectrographic speaker identification

This research was designed to investigate the effects of selected vocal disguises upon spectrographic speaker identification. The experiment consisted of open-trial, spectrogram-matching tasks with 15 ’’reference’’ and 15 ’’matching’’ speakers. The speakers produced two sentence sets in six different ways: (1) normal-speaking mode, (2) old-age disguise, (3) hoarse disguise, (4) hypernasal disguise, (5) slow-rate disguise, and (6) free disguise of his own choosing. The reference spectrograms were always undisguised speech samples; the matching spectrograms were either disguised or undisguised. Four trained spectrographic examiners completed the matching tasks. The results revealed that the examiners were able to match speakers with a moderate degree of accuracy (56.67%) when there was no attempt to vocally disguise either utterance. The inclusion of disguised speech samples in the matching tasks significantly interfered with speaker-identification performance. However, certain voice disguises were more effective than others. In addition, certain speakers were considerably more difficult to identify than others. In summary, this investigation did not substantiate prior claims that spectrographic speaker identification is unaffected by attempts at disguising one’s voice. Subject Classification: [43]70.65, [43]70.40.