Preprocessing of audio data for voice transcription systems

The object of study is the audio file transcription system. The article presents a study on the impact of using high-pass, low-pass, and bandpass filters on the process of transcribing audio files. The aim of the work is to speed up the transcription of groups of audio files. The study identifies several types of audio data for which pre-filtering is applied. The research is part of solving the problem of transcribing audio files using neural network-based systems. The article provides an overview of works on pre-processing of voice files, analyzes the advantages and disadvantages of the described approaches. The use of complex preprocessing algorithms improves the recognition quality but significantly slows down its speed or requires additional computing power. Therefore, pre-filtering is used to improve speech recognition quality without reducing the speed of the system as a whole. The article includes experimental studies on the effect of filtering frequencies on the speed of voice transcription. The obtained research results allow us to conclude that the use of a bandpass filter with a lower passband frequency in the range of 150-200 Hz and an upper passband frequency in the range of 3500-7000 Hz allows for an increase in transcription speed not only through the use of video cards but also through the use of central processors and pre-filtering. It is also proposed to remove empty segments by energy of the signal and speed up the voice, which affects the time of its transcription, using the aforementioned filters. Ref. 5, pic. 2, tabl. 7