Adapting acoustic models to new domains and conditions using untranscribed data

This paper investigates the unsupervised adaptation of an acoustic model to a domain with mismatched acoustic conditions. We use techniques borrowed from the unsupervised training literature to adapt an acoustic model trained on the Wall Street Journal corpus to the Aurora-2 domain, which is composed of read digit strings over a simulated noisy telephone channel. We show that it is possible to use untranscribed in-domain data to get significant performance improvements, even when it is severely mismatched to the acoustic model training data.