Automatic Speaker Segmentation using Multiple Features and Distance Measures: A Comparison of Three Approaches

Abstract

This paper addresses the problem of unsupervised speaker change detection. Three systems based on the Bayesian Information Criterion (BIC) are tested. The first system investigates the AudioSpectrumCentroid and the AudioWaveformEnvelope features, implements a dynamic thresholding followed by a fusion scheme, and finally applies BIC. The second method is a real-time one that uses a metric-based approach employing the line spectral pairs and the BIC to validate a potential speaker change point. The third method consists of three modules. In the first module, a measure based on second-order statistics is used; in the second module, the Euclidean distance and T 2 Hotelling statistic are applied; and in the third module, the BIC is utilized. The experiments are carried out on a dataset created by concatenating speakers from the TIMIT database, that is referred to as the TIMIT data set. A comparison between the performance of the three systems is made based on t-statistics.

Computationally Efficient and Robust BIC-Based Speaker Segmentation

“…RCL and M DR are also improved with respect to the two remaining systems. Finally, the superiority of the proposed system against the three systems developed in [21] is demonstrated by the fact that its F 1 value is relatively improved by 7.917%, 6.438%, and 28.007%, respectively. In [12], the used dataset was created by concatenating speaker utterances from the TIMIT database, too.…”

“…It outperforms three other systems tested on a similar dataset, created by concatenating speakers from the TIMIT database, as described in [21]. Although the dataset in [21] is substantially smaller than the conTIMIT test dataset, the nature of the audio recordings is the same enabling us to conduct fair comparisons. The performance achieved March 22, 2008 DRAFT by the previous approaches is summarized in Table XII.…”

Speaker segmentation and clustering

“…Features like the smoothed zerocrossing rate (SZCR), the perceptual minimum variance distortionless response (PMVDR), and the filterbank log-coefficients (FBLCs) are introduced in [53]. Additional features are derived from MPEG-7 audio standard such as AudioSpectrumCentroid, AudioWaveformEnvelope [7,8], AudioSpectrumEnvelope, and AudioSpectrumProjection [5,6].…”

“…The term modified power spectrum coefficients means that the power spectrum coefficients corresponding to frequencies below 62.5 Hz are replaced by a single coefficient equal to their sum [7,8,52].…”

“…The MPEG-7 standard developed by the Moving Picture Experts Group can be used to describe efficiently a speech recording [3,4]. For example, MPEG-7 lowlevel audio feature descriptors such as AudioSpectrumProjection, AudioSpectrumEnvelope [5,6], AudioSpectrumCentroid, AudioWaveformEnvelope [7,8] can be used. MPEG-7 high-level tools, such as SpokenContent, that exploit speakers' word usage or prosodic features, could also be exploited.…”