Different aspects of source information for limited data speaker verification

Abstract

Limited data speaker verification has shown its significance in practical system oriented applications. The paper shows the importance of different aspects of voice source feature for limited test data scenario. A baseline speaker verification system using conventional mel frequency cepstral co-efficients (MFCC) feature is developed and performance under limited test data condition (≤10 s) is evaluated. A parallel system based on source feature mel power difference of spectrum in subband (M-PDSS) is developed in the i-vector based speaker verification framework. Both the systems were fused at the score level for the cases of short segments of test speech, which demonstrated the importance of source feature with reduction in test data duration. A comparative study of the M-PDSS feature is then made with our earlier work using discrete cosine transform of the integrated linear prediction residual (DCTILPR) feature and then fusion of two source features M-PDSS and DCTILPR along with MFCC features is carried out. An absolute improvement of 5.19% is obtained for 2 s of test data which conveys the significance of multiple source information under limited data speaker verification as it carries different aspects of source information.

Exploring different attributes of source information for speaker verification with limited test data

“…The work on source features is extended to introduce mel power difference of spectrum in the subband (MPDSS) feature to capture the source information in a different manner. 13,14 Fusion of this feature with MFCC at the score level is also found to be an improvement. But at the same time it is noticed that the information captured by the two features, DCTILPR and MPDSS, had different aspects of source characteristics.…”

“…13,14 Depending on the type of voice (say, hard or soft), the spectral flatness of the LP spectrum also changes. If the voice is hard, then it results in rapid and complete closure of vocal folds.…”

“…11,12 An epoch extraction algorithm is applied, and using these epochs, a voiced/unvoiced decision based on maximum normalized cross-correlation is applied as in Refs. 17 and 18.…”

“…DCTILPR captures the glottal shape information of a speaker in a pitch synchronous manner. 11,12 But this does not capture the periodicity information of the signal denoting how much periodic it is. MPDSS feature is a variant of spectral flatness measure, and captures the periodicity information as the peak to dip ratio of the spectrum of a signal measures the periodicity.…”

“…Also, as the excitation source features are less dependent on the amount of phonetic content, the duration of enrollment/testing can be less. 9,10 In earlier work, 11,12 a source feature called discrete cosine transform of the integrated linear prediction residual (DCTILPR) is computed. The score level fusion of this feature with MFCC feature was an improvement over the baseline system using only MFCC under limited test data conditions.…”