Enhancing Automated English Speaking Tests: The Integration of Natural Language Processing and Machine Learning

Automated tests for the English-speaking ability have gained popularity in language education since they can reach a larger audience and reach and factored tests are more efficient as compared to the traditional ones. However, the current practice involves using traditional machine learning models or rule-based system to assess SP in speech which provide less or no accuracy in face of highly variable parameters such as fluency, pronunciation, and context where learner speaks different accents. To overcome these limitations, this study proposes the combination of NLP with LSTM networks to improve the performance of ASSEs for English speaking abilities of the learners. The proposed model utilizes NLP transcribe speech directly into text, as well as for subsequent textual analysis; LSTM helps capture temporal dependencies in the spoken language to allow efficient assessment of temporal features such as the pronunciation, syntactic fluency, and coherency of the recognized speech. Preprocessing includes filtering and segmenting speech, feature extraction that rely on both linguistic and acoustic properties and the training of LSTM to detect fluency, grammatical and pronunciation errors. The proposed model is validated based on a heterogeneous data set of English speakers, and considerable enhancement of the evaluation’s accuracy over all previous techniques is shown. Work carried out demonstrates improvements in scoring accuracy as well as more specific feedback given to learners in real time to allow for improvements in language abilities. This combination of NLP and LSTM come as the new proposed solution for the automated language assessments in English with an advantage of scalability, efficiency and high accuracy.