Automatic speech recognition method for English translators based on improved transfer learning

Aiming to solve the problems of high word error rate and long response time in current English translator speech recognition methods, an improved transfer learning based English translator speech automatic recognition method is proposed. Sample, quantify, and pre-process the speech signal of the English translator, and extract the features of the pre-processed English translator speech signal. Using the speech signal features of the English translator as input vectors and the speech recognition results of the English translator as output vectors, transfer learning is improved by removing the output layer of the base model, re-initialising an output layer and connecting it to the hidden layer of the base model, and fine-tuning the network to build an English translator speech automatic recognition model and obtain speech recognition results. Experimental results show that the word error rate of the proposed method is less than 3.5%, and the highest response time is only 7.3 ms.