Dialogue audio voiceprint automatic acquisition method based on deep learning

Abstract

The invention discloses a dialogue audio voiceprint automatic acquisition method based on deep learning. The method comprises the following steps: segmenting an audio into a plurality of audio clips each comprising a speaker through a speaker separation algorithm; converting each audio clip into characters through a speech recognition algorithm; through a template matching strategy, according to a recognition result of the matched audio clip, determining a voiceprint clip used for automatically collecting a speaker; and converting the text of the name recognized by voice recognition into pinyin, and matching the pinyin with a department information system to confirm the real name of the voiceprint. According to the method, the UISRNN used for speaker separation and the fine tuning Whisper model used for transcription are combined, automatic collection of voiceprints is achieved, the recognition efficiency and recognition accuracy of the voiceprints are improved, and a basis is provided for wide application of the voiceprint recognition technology.

Images