Foreign Accent Recognition with South Indian Spoken English
Radha Krishna Guntur, Ramakrishnan Krishnan, Vinay Kumar Mittal · 2020
The type of non-native speech is majorly influenced by the mother tongue. A study on Mother Tongue Influence (MTI) on a second language was carried out by employing Native Language Identification (NLI) approach. This included extraction of cepstral features, mainly the Mel Frequency Cep-stral Coefficients (MFCCs). Recognition of characteristic English accents of South Indian speakers belonging to the three regional accent groups corresponding to languages: Telugu, Tamil, and Kannada was investigated. A specifically developed corpus containing native speech samples of the speakers from each group, and non-native English speech of their compatriots was used. Experiments in text-independent mode using MFCC based Gaussian Mixture Models-Universal Background Model (GMM-UBM) classifier resulted in accuracies up to 91%, and an EER of 9.1%. Identification using i-vector paradigm produced an accuracy of 94%, with an EER of 6.1%. A general purpose recognition method is proposed for automatic classification of speakers based upon foreign accent.