Automatic speech recognition (ASR) technology has reached a mature level, and improving performance in data-scarce scenarios has become a key research focus. In this study, we propose a novel approach to building acoustic models by leveraging the multi-head attention mechanism to enhance various acoustic features extracted from audio data, combined with time-delay neural networks (TDNNs) using semi-orthogonal low-rank matrix factorization, resulting in significant performance improvements. To increase the robustness of the acoustic model, we introduce a small amount of data perturbation during the early stages of training. Additionally, during the decoding phase, we employ an external recurrent neural network language model (RNNLM) for rescoring, further enhancing the model’s accuracy. A comprehensive evaluation of the widely used LibriSpeech corpus shows that our method, using only half the training data, achieves word error rates (WERs) of 3.15 and 7.04 on the test_clean and test_other datasets, respectively. These results outperform models with similar architectures trained on the full dataset and demonstrate performance comparable to mainstream end-to-end models.
With the rapid development of Internet technology, a massive amount of English text has emerged, creating an urgent need to evaluate the semantic similarity of these texts across different domains. Existing research on semantic similarity faces the challenge of insufficient prior knowledge. Therefore, a new method has been proposed, called Knowledge Prior Topic Model (KPTM), for analyzing text semantic similarity. This method incorporates semantic relationship data from the large-scale English knowledge base, Probase, into the topic model and combines it with an effective keyword serialization approach to calculate the hybrid measurement of concepts and keywords for text similarity. Experimental results demonstrate that the proposed method contributes to improving semantic similarity and outperforms other models in comparative experiments.