The Use of Speech Recognition Systems to Select a Useful Signal in Noisy Speech at a Low Signal-To-Noise Ratio

The article considers two commonly used speech recognition systems – Yandex Speech. Kit and Google Speech Recognition. A comparison of these systems is carried out for the recognition of noisy speech at different signal-to-noise ratios. Noise is superimposed on speech at a level of 65 dB, differing from it in level from 20 dB to -5dB with a step of 5 dB. It is noted that preliminary training of the speech recognition system is not carried out; the voices of four speakers with a total recording duration of 100 minutes are used in the experiment. It has been experimentally revealed that at a low signal-to-noise ratio, speech recognition systems do not cope well enough, therefore, they need to be improved in order to have an acceptable recognition coefficient at a low signal-to-noise ratio.