Combining in-domain and out-of-domain speech data for automatic recognition of disordered speech

• DOI: 10.21437/interspeech.2013-324
• Publication Date: 2021-08-27T05:58:00Z
• Authors: H. Christensen; M. B. Aniol; Peter Bell; Phil D. Green; Thomas Hain; Simon King; Pawel Swietojanski
• Keywords: Feature (linguistics)

Recently there has been increasing interest in ways of using outof-domain (OOD) data to improve automatic speech recognition performance domains where only limited is available.This paper focuses on one such domain, namely that disordered for which very small databases exist, but normal can be considered OOD.Standard approaches handling use adaptation from OOD models into the target here we investigate an alternative approach with its focus feature extraction stage: used train feature-generating deep belief neural networks.Using AMI meeting and TED talk datasets, various tandem-based speaker independent systems as well maximum a posteriori adapted dependent systems.Results UAspeech isolated word task are promising our overall best system (using combination data) giving correctness 62.5%; increase 15% previously published results based conventional model adaptation.We show relative benefit varies considerably loosely correlated severity speaker's impairments.