Noisy speech commands recognition algorithm based on test spectral transformations of input signal

In practice, the result of signal recognition is degraded by noise. Training speech signals are usually noise-free, while testing speech signals are noisy. The presence of noise leads to a strong deviation of the spectra of the testing speech signals from the spectra of their standards in the training sample. Therefore, the quality of the recognition result against a background of noise drops sharply. The article proposes a trial amplification of the speech signal spectrum in the recognition process. A multiple algorithm for recognizing commands against a background of noise is compared with a single algorithm for recognizing speech commands. The problem of recognition of speech commands on the background noise is reviewed. The developed numerical algorithm of recognition is studied. The results of the experiments are reported on 11 speech commands from the TIDigits dataset.