Latent Sequence Decompositions
This work addresses speech recognition accuracy for tasks like transcription, but it appears incremental as it builds on existing sequence modeling approaches.
The paper tackles the problem of sequence decomposition with variable-length output units in speech recognition, achieving a 12.9% word error rate (WER) on the Wall Street Journal task, which improves to 9.6% WER when combined with a convolutional network encoder.
We present the Latent Sequence Decompositions (LSD) framework. LSD decomposes sequences with variable lengthed output units as a function of both the input sequence and the output sequence. We present a training algorithm which samples valid extensions and an approximate decoding algorithm. We experiment with the Wall Street Journal speech recognition task. Our LSD model achieves 12.9% WER compared to a character baseline of 14.8% WER. When combined with a convolutional network on the encoder, we achieve 9.6% WER.