Research on Modeling Units of Transformer Transducer for Mandarin Speech Recognition
This work addresses performance improvements for Mandarin speech recognition, presenting incremental advancements in modeling units and training methods.
The paper tackles improving Mandarin speech recognition by proposing a transformer transducer combining self-attention and RNN, exploring modeling units, and introducing a mix-bandwidth training method. The results show that using syllable with tone as the modeling unit achieves the best performance, with relative reductions of 14.4% and 44.1% in WER compared to other units, and a 13.5% reduction in CER.
Modeling unit and model architecture are two key factors of Recurrent Neural Network Transducer (RNN-T) in end-to-end speech recognition. To improve the performance of RNN-T for Mandarin speech recognition task, a novel transformer transducer with the combination architecture of self-attention transformer and RNN is proposed. And then the choice of different modeling units for transformer transducer is explored. In addition, we present a new mix-bandwidth training method to obtain a general model that is able to accurately recognize Mandarin speech with different sampling rates simultaneously. All of our experiments are conducted on about 12,000 hours of Mandarin speech with sampling rate in 8kHz and 16kHz. Experimental results show that Mandarin transformer transducer using syllable with tone achieves the best performance. It yields an average of 14.4% and 44.1% relative Word Error Rate (WER) reduction when compared with the models using syllable initial/final with tone and Chinese character, respectively. Also, it outperforms the model based on syllable initial/final with tone with an average of 13.5% relative Character Error Rate (CER) reduction.