Identification of a Speaker’s Gender by Voice Characteristics on the Background of Multi-Talker Noise

Psychophysical methods were used to study the features of the recognition of speakers’ gender based on voice characteristics in conditions of speech-like interference and stimulation via headphones. A set of speech signals and multi-talker noise from experiments in a free sound field, i.e., a spatial scene [Andreeva et al., 2019], were used. The set included eight disyllabic words pronounced by four speakers: two male and two female voices with mean fundamental frequencies of 117, 139, 208, and 234 Hz. Multi-speaker noise was obtained by mixing all the audio files (eight words × four speakers). The signal-to-noise ratio was 1:1, which subjectively corresponded to the maximum noise level in the spatial scene (SNR = –14 dB). A total of 42 adult subjects (17–57 years old) took part in the experiments. Additionally, three age subgroups were defined: 18.6 ± 1.5 years ( n = 27), 28 ± 4.1 years ( n = 7), and 46 ± 5.4 years ( n = 8). All subjects had normal hearing. The study results and comparison with data from cited work confirmed the importance of voice characteristics for the auditory analysis of complex spatial (free sound field) and non-spatial (headphones) scenes, and also demonstrated the role of masking mechanisms and binaural perception, particularly the high-frequency mechanism of spatial hearing. In addition, a relationship was found between the perceptual assessment of the gender characteristics of voices in noise on the one hand and subjects’ age and speakers’ gender (male or female voice) neuron the other. The results are of practical value in terms of organizing speech-related hearing training, in the early diagnosis of disorders of noise immunity in hearing speech, and the development of noise-immune systems for automatic speaker verification and hearing aid technologies.