Suitability of speech files for automatic speech recognition systems after noise reduction procedures

R. Kh. Latypov^a, E. L. Stolov<sup>b

a</sup> Institute of Computer Mathematics and Information Technologies, Kazan (Volga Region) Federal University, Kazan, Russia
^b Institute of Computer Mathematics and Information Technologies, Kazan (Volga Region) Federal University, Kazan, Russia

Abstract: The results of the experiments on noise reduction in speech files with further transfer in the Google automatic speech recognition (ASR) system are presented. An original method is developed to identify the intervals containing only noise. It is shown that using a modified Wiener filter in the time domain allows to improve the recognition quality of noisy files.

Keywords: speech file, noise reduction, speech recognition, Google, Wiener filter.

UDC: 534.4+004.93

Received: 26.08.2015

Bibliographic databases:

•