Towards Effective and Compact Contextual Representation for Conformer Transducer Speech Recognition Systems
This addresses the challenge of improving streaming speech recognition accuracy by efficiently using historical context, though it is incremental as it builds on existing Conformer-Transducer methods.
The paper tackled the problem of incorporating long-range cross-utterance context in ASR systems by learning compact low-dimensional contextual features in a Conformer-Transducer Encoder, resulting in statistically significant WER reductions of 0.7% to 0.5% absolute (4.3% to 3.1% relative) on dev and test data.
Current ASR systems are mainly trained and evaluated at the utterance level. Long range cross utterance context can be incorporated. A key task is to derive a suitable compact representation of the most relevant history contexts. In contrast to previous researches based on either LSTM-RNN encoded histories that attenuate the information from longer range contexts, or frame level concatenation of transformer context embeddings, in this paper compact low-dimensional cross utterance contextual features are learned in the Conformer-Transducer Encoder using specially designed attention pooling layers that are applied over efficiently cached preceding utterances history vectors. Experiments on the 1000-hr Gigaspeech corpus demonstrate that the proposed contextualized streaming Conformer-Transducers outperform the baseline using utterance internal context only with statistically significant WER reductions of 0.7% to 0.5% absolute (4.3% to 3.1% relative) on the dev and test data.