Multi-head Monotonic Chunkwise Attention For Online Speech Recognition
This work addresses the need for efficient online speech recognition systems, offering incremental improvements over existing methods for applications like real-time voice interfaces.
The paper tackled the problem of online speech recognition by proposing multi-head monotonic chunk-wise attention (MTH-MoChA), an improved version of MoChA that splits input sequences into chunks for multi-head attention, achieving a character error rate reduction from 8.96% to 7.68% on AISHELL-1 and 7.28% on a large in-car dataset.
The attention mechanism of the Listen, Attend and Spell (LAS) model requires the whole input sequence to calculate the attention context and thus is not suitable for online speech recognition. To deal with this problem, we propose multi-head monotonic chunk-wise attention (MTH-MoChA), an improved version of MoChA. MTH-MoChA splits the input sequence into small chunks and computes multi-head attentions over the chunks. We also explore useful training strategies such as LSTM pooling, minimum world error rate training and SpecAugment to further improve the performance of MTH-MoChA. Experiments on AISHELL-1 data show that the proposed model, along with the training strategies, improve the character error rate (CER) of MoChA from 8.96% to 7.68% on test set. On another 18000 hours in-car speech data set, MTH-MoChA obtains 7.28% CER, which is significantly better than a state-of-the-art hybrid system.