Read, Attend and Pronounce: An Attention-Based Approach for Grapheme-To-Phoneme Conversion

We propose an attention-enabled encoder-decoder model for the problem of grapheme-to-phoneme conversion. Most previous work has tackled the problem via joint sequence models that require explicit alignments for training. In contrast, the attentionenabled encoder-decoder model allows for jointly learning to align and convert characters to phonemes. With this approach, we achieve state-of-the-art results on the CMUDict data set with a Word Error Rate (WER) of 23.62%. The performance is comparable even to that of models trained using aligned training data.