Amit S. Malegaonkar

Efficient Speaker Change Detection Using Adapted Gaussian Mixture Models

A new approach to speaker change detection is proposed and investigated. The method, which is based on a probabilistic framework, provides an effective means for tackling the problem posed by phonetic variation in high-resolution speaker change detection. Additionally, the approach incorporates the capability for dealing with undesired effects of variations in speech characteristics. Using the experimental investigations conduced with clean and broadcast news audio, it is shown that the proposed method is significantly more effective than the currently popular techniques for speaker change detection. To enhance the computational efficiency of the proposed method, modified implementation algorithms are introduced which are based on the exploitation of the redundant operations and a fast scoring procedure. It is shown that, through the use of the proposed fast algorithm, the computational efficiency of the approach can be increased by over 77% without significant reduction in its accuracy. The paper discusses the principles and characteristics of the proposed speaker change detection method, and provides a detailed description of its efficient implementation. The experiments, investigating the performance of the proposed method and its effectiveness in relation to other approaches, are described and an analysis of the results is presented.

A test of the effectiveness of speaker verification for differentiating between identical twins

This paper presents investigations into the ability of speaker verification technology to discriminate between identical twins. It is shown that whilst, in general, the genetic and non-genetic characteristics of voice are both of value to speaker verification capabilities, it is the latter which is highly beneficial in the separation of the speech of identical twins. It is further demonstrated that through the use of unconstrained cohort normalisation as a complementary means for the exploitation of such voice characteristics, the verification reliability can be considerably enhanced for both identical twins and unrelated speakers. Experiments were conducted using a bespoke clean-speech database consisting of utterances from 49 identical twin pairs. The paper details the problem in speaker verification posed by identical twins, discusses the experimental investigations and provides an analysis of the results.

Unsupervised speaker change detection using probabilistic pattern matching

This letter presents an investigation into the use of a probabilistic pattern matching approach for detecting speaker changes in audio streams. The experiments are conducted using clean speech as well as broadcast news material. It is shown that, in the proposed approach, the use of bilateral scoring is considerably more effective than unilateral scoring. Appropriate score normalization methods are considered in the study. It is observed that in all the cases, the bilateral scoring approach outperforms the currently popular method of Bayesian information criterion (BIC) for speaker change detection. This letter discusses the principles of the proposed approach and details the experimental investigations

Qualitative fusion of normalised scores in multimodal biometrics

A new approach to enhancing the accuracy of multimodal biometrics is investigated. The proposed approach, which involves combining score normalisation and qualitative-based fusion, is shown to considerably improve the accuracy of multimodal biometrics under different data conditions.

Performance evaluation in open-set speaker identification

The concern in this study is the approach to evaluating the performance of the open-set speaker identification process. In essence, such a process involves first identifying the speaker model in the database that best matches the given test utterance, and then determining if the test utterance has actually been produced by the speaker associated with the best-matched model. Whilst, conventionally, the performance of each of these two sub-processes is evaluated independently, it is argued that the use of a measure of performance for the complete process can provide a more useful basis for comparing the effectiveness of different systems. Based on this argument, an approach to assessing the performance of open-set speaker identification is considered in this paper, which is in principle similar to the method used for computing the diarisation error rate. The paper details the above approach for assessing the performance of open-set speaker identification and presents an analysis of its characteristics.

On the enhancement of speaker identification accuracy using weighted bilateral scoring

This paper presents investigations into an effective bilateral scoring method in open-set speaker identification. The approach is based on the fact that two different speakers usually are not reciprocal. A difficulty in deploying bilateral scoring is that test utterances are normally much shorter than training utterances. To tackle this problem, the proposed approach provides the final identification score based on a weighted combination of independently normalised forward and reverse scores. Based on the experimental results obtained using clean and telephone quality speech, it is shown that the proposed approach is more effective than the conventional scoring methods in open-set speaker identification.

Enhancement of multimodal biometric segregation using unconstrained cohort normalisation

This paper presents an investigation into the effects, on the accuracy of multimodal biometrics, of introducing unconstrained cohort normalisation (UCN) into the score-level fusion process. Whilst score normalisation has been widely used in voice biometrics, its effectiveness in other biometrics has not been previously investigated. This study aims to explore the potential usefulness of the said score normalisation technique in face biometrics and to investigate its effectiveness for enhancing the accuracy of multimodal biometrics. The experimental investigations involve the two recognition modes of verification and open-set identification, in clean mixed-quality and degraded data conditions. Based on the experimental results, it is demonstrated that the capabilities provided by UCN can significantly improve the accuracy of fused biometrics. The paper presents the motivation for, and the potential advantages of, the proposed approach and details the experimental study.