Speech recognition method and device

Abstract

The invention discloses a method and a device for voice recognition. The method comprises the steps of: obtaining voice data to be recognized, and extracting voice characteristics; decoding the voice characteristics simultaneously according to a first linguistic model and a second linguistic model, obtaining a first voice recognition result, a second voice recognition result, a first cumulative probability under the first voice linguistic model and a second cumulative probability under the second voice linguistic model; determining a final recognition result according to the first cumulative probability and the second cumulative probability; and judging whether to carry out the voice recognition for the voice data to be recognized according to the provenance of the final recognition result. By adopting the voice recognition method, a stable and reliable voice recognition environment is provided and smoothness of human-computer interaction is ensured.

Description

technical field [0001] The present invention relates to the technical field of speech recognition, and in particular, to a speech recognition method and device. Background technique [0002] With the development of speech recognition technology and the continuous improvement of recognition rate, speech recognition technology is more and more applied in various scenes of daily life. Among them, dialogue interaction is a very important field. Users can communicate with the computer through voice, so that the computer can understand the user's intention and complete the user's instructions. A complete interaction process may include several dialogue scenarios, such as the "hotel reservation" process, including the following scenarios: first tell the computer the city of the hotel booked; secondly, check-in time; then room type and number; lastly, check out time. From the first scene, the computer recognizes the user's voice, judges the user's intention, and jumps to the next ...

AI Technical Summary

Problems solved by technology

[0004] However, the existing problems are: (1) for the above-mentioned first method, the dynamic range of the ppl value is relatively large, which is an unbounded value, and it is difficult to determine a suitable threshold for the unbounded value, which may easily cause false alarms and missed detections. Moreover, although some recognition results are wrong, the ppl value of the recognition result may be relatively high, resulting in a judgment error; (2) For the second method above, because the probability distribution of the language model is unbalanced, the recognition result will occur. At the same time as the bias, it also makes it difficult for the word network to fully include effective competing candidates, resulting in inaccurate calculated confidence, resulting in wrong judgments, thereby reducing the stability and reliability of the speech recognition system, and because of the confidence, the recognition cannot be judged Whether the result conforms to the mode required by the scene context, so the recognition result judged to be correct according to the confidence level may also block the interaction process because it does not meet the scene requirements

Images