This study explores some options for improving the performance of F0-based speaker verification. We tested different parameterisation techniques that enable us to capture non-unimodal distribution. We also tested the use of dynamic features (delta F0) and different scales (log10). As a result, we discovered that combinations of these techniques could significantly improve both the performance of speaker verification, and the reliability of the likelihood ratios.
|Title of host publication||Proceedings of the 13th Australasian International Conference on Speech Science and Technology|
|Place of Publication||Melbourne, Australia|
|Publisher||Australian Speech Science and Technology Association (ASSTA)|
|Number of pages||4|
|Publication status||Published - 2010|
|Event||SST 2010: Thirteenth Australasian International Conference on Speech Science and Technology 2010 - Melbourne, Australia|
Duration: 14 Dec 2010 → 16 Dec 2010
|Conference||SST 2010: Thirteenth Australasian International Conference on Speech Science and Technology 2010|
|Period||14/12/10 → 16/12/10|
Kinoshita, Y., Ishihara, S., & BAKKER, D. (2010). F0 can tell us more: speaker verification using the long term distribution. In Proceedings of the 13th Australasian International Conference on Speech Science and Technology (pp. 50-53). Melbourne, Australia: Australian Speech Science and Technology Association (ASSTA).