Speech recognition correction method, corresponding device, equipment and medium

Abstract

The invention discloses a speech recognition correction method, a corresponding device, equipment and a medium. The method comprises the following steps: acquiring a preliminary audio text and confidence data recognized from original audio data by a selected acoustic model; replacing words with confidence lower than a preset threshold value in the preliminary audio text with hole marks to obtain a marked audio text; performing text alignment on the marked audio text according to an original audio text of the original audio data, so that the hole marks in the marked audio text are correspondingly complemented according to the original audio text to obtain a corrected audio text; and marking the original audio data as a training sample, marking the corrected audio text as a supervision label of the original audio data, and storing the training sample and the supervision label in a sample library required by acoustic model training. According to the method and the device, dirty data formed by the audio text and the audio data associated with the same voice content can be efficiently cleaned, so that the training data required by acoustic model training can be prepared.

Description

technical field [0001] The embodiments of the present application relate to the technical field of speech recognition, and in particular, to a speech recognition correction method and corresponding devices, equipment, and media. Background technique [0002] A large amount of training data is precisely the essential material for an excellent acoustic model. At present, the mainstream method of producing ASR (Automatic Speech Recognition, automatic speech recognition) training data is often obtained through direct sampling: by manually reading a certain text or dialogue accurately, Thus, the audio data formed by reading aloud and the audio text read are obtained. The audio file can be used as a training sample, and its audio text can be used as a supervision label, so high-quality training corpus can be produced, which can be directly used for training. data. Obviously, in this method, the collection efficiency is very low, and the data acquisition cost is very high. [000...

AI Technical Summary

Problems solved by technology

However, the audio text part of these data often loses its timing correspondence with the audio data. Generally, this kind of data is named "dirty data", which cannot be directly used for the training of the acoustic model, so it needs to be modified. Further processing in order to produce useful training samples, so the key to the problem is how to construct an effective technical solution to achieve efficient production of acoustic model training samples

Images