Speech recognition and enhancement in underground mines for the use of smart voice assistants

Abstract

Smart voice assistants are being used progressively in new fields and environments. This trend has also reached Polish underground mines, where the first attempts to implement solutions of this type are already being executed. Unfortunately, the underground mine environment places great demands on the operation of such assistants. This problem is even greater when we take into account less popular and more difficult languages like Polish. In this article, we approach the problem of noise in the underground mine environment and the correct transmission of speech to text by a smart voice assistant. In particular, we researched the possibility of using various algorithms for speech enhancement. This was achieved by taking a series of recordings in an underground environment, during which a voice assistant was used in a standardized way by different people. This voice material was then denoised using 10 different algorithms, converted to text, and evaluated. Such a comparison showed that only one of the chosen methods influenced the speech-to-text solution in a positive way, showing that current algorithms are not sufficient for harsh industrial environments similar to underground mines.