Abstract
General purpose automatic speech recognition
(gpASR) systems such as Google, Watson, etc.
sometimes output inaccurate sentences when used
in a domain specific scenario as it may not have had
enough training samples for that particular domain
and context. Further, the accent of the speaker and
the environmental conditions in which the speaker
speaks a sentence may influence the speech engine
to recognize certain words inaccurately. Many approaches
to improve the accuracy of ASR output
exist. However, in the context of a domain and
the environment in which a speaker speaks the sentences,
gpASR output needs a lot of improvement
in order to provide effective speech interfaces to
domain-specific systems. In this paper, we demonstrate
a method that combines bio-inspired artifi-
cial development (ArtDev) with machine learning
(ML) approaches to repair the output of a gpASR.
Our method factors in the environment to tailor the
repair process.
Keywords
- Natural Language Processing: Speech
- Machine Learning: Machine Learning
- Natural Language Processing: Natural Language Processing
- Knowledge Representation, Reasoning, and Logic: Reasoning about Knowlege and Belief
