Pilot Experiments on Children's Voice Recording

摘要

Automatic speech recognition is being used increasingly in a variety of applications. There is great potential for its use in educational applications for children. However, the accuracy of recognition of child speech is very low. There are probably a number of reasons for this, but one is the difficulty in collecting high-quality recordings of children to be used in the building of speech models. If a better interface can be provided between the child and the recording equipment then it may be possible to collect better samples. Interfaces have been designed to be tested to that end, using alternative interface paradigms: push-to-talk and a limited time recording with and without a progress bar. These alternatives will be compared by collecting speech samples and measuring their quality.