Daniel J. Mashao

Rapid and brief communication: Combining classifier decisions for robust speaker identification

In this work, we combine the decisions of two classifiers as an alternative means of improving the performance of a speaker recognition system in adverse environments. The difference between these classifiers is in their feature-sets. One system is based on the popular mel-frequency cepstral coefficients (MFCC) and the other on the new parametric feature-sets (PFS) algorithm. The feature-vectors both have mel-scale spectral warping and are computed in the cepstral domain but the feature-sets differs in the use of spectral filters and compressions. The performance of the classifier is not much different in recognition rates terms but they are complementary. This shows that there is information that is not captured in the popular mel-frequency cepstral coefficients (MFCC), and the parametric feature-sets (PFS) is able to add further information for improved performance. Several ways of combining these classifiers gives significant improvements in a speaker identification task using a very large telephone degraded NTIMIT database.

Modified Segmental Histogram Equalization for robust speaker verification

Abstract It is well known that when there is an acoustic mismatch between the speech obtained during training and testing the accuracy of speaker recognition systems drastically deteriorates. In this paper we propose Modified Segmental Histogram Equalization to improve the robustness of a speaker verification system operating in telephone environments. The technique transforms the features extracted from short adjacent segments of speech within an utterance such that their statistics conform to that of a Gaussian distribution with zero mean and unity variance across all recording conditions. In doing so, the feature statistics become less environment-dependent. Experiments performed on the NIST 2000 database show significant improvements in performance.

Usability Engineering of an Interactive Voice Response System in a Diverse-Cultured and Multilingual Setting

This paper advocates the utilization of user centred approach in the design of the user interface (UI) of a speech-enabled interactive voice response (IVR) system for the provision of information to the diverse user population of Southern Africa. It is postulated that the end-product of this research will allow users to access information through a telephone line in their language of choice amongst the various official languages of South Africa and will therefore benefit all the citizens of the country including those with low literacy levels and little or no experience with contemporary ICT

A hybrid GMM-SVM speaker identification system

This paper proposes a system that combines the power of generative Gaussian mixture models (GMM) and discriminative support vector machines (SVM) in a speaker identification task. The classification methods are different and they also exhibit uncorrelated errors and this is used to improve performance of the speaker identification system. Whereas GMM needs more data to perform adequately and is computationally inexpensive, SVM on the other hand can do well with less data and is computationally expensive. A system where SVM post processes the results of a GMM system is proposed and it is shown that it is able to reduce speaker identification errors by over 11% on a database with 630 speakers. Similar hybrid systems have been proposed before but this is unique since both classifiers use the same feature vectors. Improved performance is found by using optimal parameters (sigma, C) for the SVM Gaussian kernel

Dynamic bandwidth allocation protocol for hierarchically compressed video in ATM networks

In this paper, a dynamic bandwidth allocation algorithm based on renegotiation of the QoS parameters of scalable hierarchical coded video is proposed. The VBR video is very bursty in nature as result of the encoding technique and the peak bit rate value could become very important compared to the average bit rate. To control the burstiness of such applications a renegotiation process of the applications QoS with the network is needed. The main idea in our algorithm is the use of hierarchical multilevel QoS profile for those applications flexible enough to accept dynamic adjustment of their QoS requirement during connection lifetime. The online predictors are used to forecast the future required bandwidth. In our proposed algorithm, even if the prediction process for more resources cannot be provisioned, the application can gracefully degrade its QoS and consequently the bandwidth requirement based on the currently available resources.

Performance evaluation of dynamic bandwidth allocation scheme for VBR video streaming in ATM networks

In this paper, we study the problem of bandwidth allocation for ATM network loaded with real-time VBR compressed video traffic. The compressed video traffic is highly bursty in nature as a result of encoding technique and shows time-variant statistical characteristics due to scene changes. These characteristics make it more difficult to manage network recourses and can cause a significant reduction in network utilization. Our proposed scheme adjusts the allocated bandwidth at regular intervals using a multilayer compressed video in the context of congestion control. The dynamic bandwidth allocation scheme adds and drops layers of the video stream during a renegotiation process to react to congestion on very short timescales. Our simulation results show a significant improvement on the performance of the proposed dynamic bandwidth allocation scheme compared with traditional static bandwidth allocation schemes, and also offers several advantages over static bandwidth allocation schemes including it removes the need of accurate traffic parameters declaration by users at call setup phase, it is valid for a wide variety of traffic sources; and it adapts quickly to the variation of traffic sources with small renegotiation intervals.

Impact of renegotiation frequency on ATM network performance

Bandwidth requirements for variable bit rate (VBR) traffic such, as compressed video, shows a dramatic variation over different time intervals. Allocating a bandwidth at peak bit rate for the entire call duration will guarantee the quality of service (QoS) but this result in poor network resources utilization. We have proposed an alternative approach that renegotiates the allocated bandwidth and QoS during the lifetime of the connection. In this paper, we present the results of a performance evaluation for the proposed renegotiation based dynamic bandwidth allocation scheme. In particular, we determine the impact of the renegotiation frequency on the network performance.

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 parameterised feature sets (PFS) to form speaker models. During testing, the likelihood of a talker, given a set of speaker models is measured. The performance of the 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 errors obtained from the PFS and the final identification is performed on smaller population. Results obtained using 2-best list show performance improvement

Quality of service renegotiation effect on ATM network performance

The bandwidth requirements for variable bit rate (VBR) traffic such as compressed video shows a dramatic variation over different time intervals. Allocating a bandwidth at the peak bit rate for the entire call duration will guarantee the quality of service (QoS) but this result in poor network resources utilization. We have proposed an alternative approach that renegotiates the allocated bandwidth and QoS during the lifetime of the connection. We present the results of a performance evaluation for the proposed renegotiation-based dynamic bandwidth allocation scheme. In particular, we determine the impact of the renegotiation frequency on the network performance.