The influence of the number of mel-frequency cepstral coefficients on the voice recognition error probability by a distributed fiber optic sensor

The study explores the potential applications of fiber optic sensors for user verification tasks. An experimental validation of the considered methods for recognizing speech signals obtained from acoustic sensors is conducted. In the experiment, voices of four different speakers were recorded. Each speaker sequentially uttered the same text simultaneously using a capacitor microphone and a fiber optic distributed sensor. To create the dataset, each speaker read three prepared texts. After training the system on the dataset and achieving speech signal recognition with sufficiently high accuracy recorded from the capacitor microphone, attention was turned to addressing the challenge of noise elimination. For this purpose, background noise was separately recorded at the experimental site. The conclusion is drawn regarding the applicability of mel-frequency cepstral coefficients for solving the task of identifying individuals based on the sound signal obtained from the fiber optic sensor with a specified accuracy.