Solving the problem of speech commands recognition

In terms of technology development, speech recognition has a long history marked by several waves of major innovations. More recently, the field has been boosted by advances in deep learning and big data. These advances are evidenced not only by the growing number of scientific papers published in this area, but by the worldwide adoption of various deep learning methods in the design and implementation of speech recognition systems. The wide variety of speech signal processing tasks, as well as its high variability and instability of processing results in general, require a new formulation of the processing task in this area. Given staging tasks identification models speech production with purpose adequate perception. Researched solution tasks speech recognition teams. Tasks identification models of speech production with purpose adequate perception are presented and two adjustment schemes for speakers to improve speech recognition out signals. The transformation of speech signals and their recognition are implemented using likelihood functions. Given results experiments with proposed schemes under-construction sites. One hundred experiments were carried out on speech signals from the public access database TIDigits 1.0.