Short Phrase Speaker Identification in Noisy Environment
Petr V. Malinin, Viktor V. Polyakov · 2018 XIV International Scientific-Technical Conference on Actual Problems of Electronics Instrument Engineering (APEIE) · 2018
The article deals with the problem of speaker identification, which uses short phrases in noisy environment. The speaker identification in harsh conditions (short phrases, noisy environment) is an actual task, since it is in harsh conditions that the possibilities of the methods used in identification are manifested. Short phrases are characterized by a minimum content of informative elements (phonemes). Usually the duration of short phrases varies hundreds of milliseconds up to several seconds, which leads to certain difficulties in identifying a person by voice. As a method of extracting voice attributes, it is proposed to use the Morlet wavelet transforms. The speakers identification results obtained on basis of the k-nearest neighbors method are presented. Identification results were obtained on the basis of voice records from the database “Acoustic speech signals for the identity system by voice data.” Noisy conditions were formed by additive overlap of different noise levels on voice records. The most important types of noise in real conditions were used white noise, speech-like noise and street noise. The approach proposed is recommended to be used in identification systems with access control of information security.