Performance of the echo canceller can be conventionally measured by putting white Gaussian noise into the echo canceller system. However, white Gaussian noise is not adequate as the test signal, since the performance may depend on the characteristics of input test signal, and the characteristics of the white Gaussian noise differ from those of real spoken language. This paper describes characteristics of spoken language required for objective quality evaluation of echo cancellers.
Following test signals having various time and frequency characteristics of spoken language are examined: white Gaussian noise, frequency weighted Gaussian noise, artificial voice, composite source signal, and real voice.
It is concluded that artificial voice having average characteristics of spoken language in time and frequency domain is satisfied for objective quality evaluation of echo cancellers other than conventional white Gaussian noise, frequency weighted white Gaussian noise, and composite source signal.
Cite as: Kitawaki, N., Asano, F., Yamada, T. (2000) Characteristics of spoken language required for objective quality evaluation of echo cancellers. Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000), vol. 3, 1101-1104, doi: 10.21437/ICSLP.2000-728
@inproceedings{kitawaki00_icslp,
author={Nobuhiko Kitawaki and Futoshi Asano and Takeshi Yamada},
title={{Characteristics of spoken language required for objective quality evaluation of echo cancellers}},
year=2000,
booktitle={Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000)},
pages={vol. 3, 1101-1104},
doi={10.21437/ICSLP.2000-728}
}