PronouncePerfect: Speech mispronunciation detection and correction system
N Meghana, Namita Hansraj Patil, M. Jahnavi, Mahimaa CS, P. Preethi · 2025
PronouncePerfect is designed to detect and correct mispronunciations in spoken English, specifically in an American accent. As pronunciation of words can vary significantly across various regions of the world such as in Britain, Australia, and America, non-native speakers can find it challenging to master pronunciation in a specific accent. The main difficulty in learning American English is accurately pronouncing phonemes, which are distinct sounds that help differentiate words. Some phonemes may not even exist in the speaker’s native language, hence they may not be able to accurately pronounce a word leading to misunderstandings and it can also hamper fluency in communication. This system is designed to not only support non-native English speakers but also individuals with speech impairments. This research, leverages advanced speech processing technologies to minimize mispronunciations in everyday speech.To achieve this, the system uses a Bidirectional Long Short-Term Memory (BiLSTM) model for mispronunciation detection, which achieves an accuracy of 90.08%. Additionally, another BiLSTM model is used for correction, with an accuracy of 72.0%. The system integrates speech-to-text conversion, phonetic transcriptions from the text, detection and correction of mispronunciations. PronouncePerfect delivers real-time and personalized feedback to help users improve their pronunciation. PronouncePerfect could be an accessible and valuable tool in both language learning and speech therapy.