Modular Domain Adaptation for Conformer-Based Streaming ASR
This work addresses the challenge of efficient domain adaptation in streaming automatic speech recognition, offering a modular solution to avoid full retraining, though it is incremental as it builds on existing Conformer transducer architectures.
The paper tackled the problem of retraining multidomain speech recognition models when domain data changes, proposing a modular domain adaptation framework that enables a single model to handle multiple domains with domain-specific parameters. Experimental results on a streaming Conformer transducer showed that the MDA-based model achieved similar performance to a multidomain model on domains like voice search and dictation by adding per-domain adapters and feed-forward networks.
Speech data from different domains has distinct acoustic and linguistic characteristics. It is common to train a single multidomain model such as a Conformer transducer for speech recognition on a mixture of data from all domains. However, changing data in one domain or adding a new domain would require the multidomain model to be retrained. To this end, we propose a framework called modular domain adaptation (MDA) that enables a single model to process multidomain data while keeping all parameters domain-specific, i.e., each parameter is only trained by data from one domain. On a streaming Conformer transducer trained only on video caption data, experimental results show that an MDA-based model can reach similar performance as the multidomain model on other domains such as voice search and dictation by adding per-domain adapters and per-domain feed-forward networks in the Conformer encoder.