Voice keyword recognition method based on end-to-end, device thereof and equipment

Abstract

The invention discloses a voice keyword recognition method based on end-to-end, a device thereof and equipment. The concept of the invention is to combine with an end-to-end thought; according to themethod, a pre-built keyword recognition network is directly fitted from the features to the target, so that the recognition process is simpler and more efficient, the superposition effect of adverse effects can be avoided, meanwhile, the keyword recognition network is easier to achieve global optimization, the development cost can be effectively reduced, and therefore, the method has a higher practical value in an actual business scene. According to the invention, an acquisition strategy of identification features is improved; therefore, the pronunciation characteristics adapting to the business scene can be fully represented; therefore, more potential key information can be captured; in addition, the keyword recognition network architecture provided by the invention can utilize the context information from the acoustic perspective, so that the defect that the existing scheme only performs recognition through an isolated pronunciation sample is fundamentally overcome, and the processing effect of locking the keywords from the audio is further obviously improved.

Description

technical field [0001] The present invention relates to the field of speech processing, in particular to an end-to-end speech keyword recognition method, device and equipment. Background technique [0002] The speech key word recognition (Keyword Spotting) that the present invention pays close attention to specifically refers to the technology of identifying and judging the content and position of a specific keyword in a given speech through pronunciation similarity under the condition of no corresponding language speech recognizer, That is to say, the technology of transcribing speech into text by a speech recognizer and then retrieving keywords from the text is not a concept. Speech keyword recognition technology is a technology between continuous speech recognition and isolated word recognition. In a given continuous, unrestricted natural speech stream, it can be recognized from the acoustic level whether a speech contains a given word or not. keywords, and give the star...

AI Technical Summary

Problems solved by technology

[0003] However, the current solution for speech keyword recognition relies on the processing effects of many independent tasks, which have a large mutual influence and are easy to magnify and superimpose errors. From feature selection to recognition modeling, the simple phoneme classification idea and real speech The keyword recognition scene does not match, and can only rely on pronunciation similarity for target recognition

Due to the above defects, the current technical means for speech keyword recognition have problems such as low processing efficiency, poor overall performance, too many false alarms and unsatisfactory recognition results.

Images