Human and Automatic Speaker Recognition over Telecommunication Channels by Laura Fernández Gallardo

Author:Laura Fernández Gallardo
Language: eng
Format: epub
Publisher: Springer Singapore, Singapore

G.722

G.722

2.01

1.65

0.008

AMR-WB

AMR-WB

2.24

1.99

0.008

Table 5.6 summarises the results in the case of no mismatch. The MixedBW system offers better performance than the bandwidth-dependent systems except for the evaluation with the G.711 codec (this difference is not statistically significant). Besides, for all the evaluations, the HTER is significantly lower than for the corresponding codec-dependent systems evaluated with the same codec (Table 5.3). For the MixedBW system, the performances of the two AMR codecs are comparable, while the rest of differences comparing NB to WB transmissions are statistically significant.

The benefits of this outcome are that a distortion-independent system would be able to perform reasonably well with all codecs despite the differences in bandwidth and that there is no need to develop codec-dependent systems or to detect the codec applied to the test segments. It can be hypothesised that a JFA system built incorporating even more channel variability in the training set would outperform the MixedBW system for the same evaluations and that the most advantageous number of eigenchannels can be adjusted.Table 5.7Performance of the bandwidth-dependent and of the distortion-independent JFA systems under codec mismatch between enrol and test segments

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.