Systems and methods for multiple speaker speech recognition

A speech recognition system, comprising:

• at least one storage device storing a set of instructions;
• and
• at least one processor in communication with the at least one storage device, wherein when executing the set of instructions, the at least one processor is configured to cause the system to perform operations including:
• obtaining speech data, the speech data including speech of a plurality of speakers;
• determining feature information of the speech data;
• obtaining a plurality of candidate speech recognition results and an initial confidence score of each of the plurality of candidate speech recognition results by processing the feature information using a decoding model;
• for each of the plurality of candidate speech recognition results,
• determining a correction value of the candidate speech recognition result using a re-scoring model;
• and
• determining the confidence score of the candidate speech recognition result by summing the initial confidence score and the correction value of the candidate speech recognition result;
• and
• determining a speech recognition result from the plurality of candidate speech recognition results based on the confidence scores of the plurality of candidate speech recognition results, the speech recognition result including a plurality of words, wherein the re-scoring model is generated based on a process of training model, the process of training model including:
• obtaining a first language model and a second language model, and the first language model being obtained by cropping the second language model;
• generating the decoding model based on the first language model;
• generating a first scoring model and a second scoring model based on the first language model and the second language model;
• and
• obtaining the re-scoring model based on the first scoring model and the second scoring model, wherein the obtaining the re-scoring model based on the first scoring model and the second scoring model includes:
• obtaining a plurality of possible speech recognition results and a first confidence score of each of the plurality of possible speech recognition results by traversing the second scoring model;
• for each of the plurality of possible speech recognition results, determining a second confidence score of the possible speech recognition result based on the first scoring model;
• updating the second scoring model based on the first confidence scores and the second confidence scores of the plurality of possible speech recognition results;
• and
• determining the re-scoring model based on the updated second scoring model.