Periodic Reporting for period 1 - MorSR (Integrating morpho-phonology in speech recognition)
Reporting period: 2019-03-01 to 2020-08-31
Development on the MorSR project has produced several working prototypes. A pronunciation app prompts second-language learners to produce words or sentences in their target language and provides tailored individual feedback on their mispronunciations. This pronunciation training system is also currently undergoing integration with the online language learners’ dictionaries of a major publishing corporation, providing wider access for language learners to develop and improve their English pronunciation skills. Additionally, the underlying speech recognition model has been developed into command recognition systems enabling recognition of specified words and phrases in novel languages on which the model has had no original training (Hindi and Bengali). Research continues on the possibility of recognising phonological features (the underlying building blocks of speech sounds) in the context of novel speech sounds that the model has not been trained on. For example, testing indicates some success in transferal from training on the [nasal] component of the English nasal consonants to detection of the same feature in nasal and nasalised vowels in Bengali.