Part of major project requirement for my undergraduate degree.
This project aims at identifying a speaker's regional language English accent from within a predetermined set of Indian accents. The project was further extended to recognize the speaker from a database of speakers where speakers can be enrolled. The frame- work has been discussed in the report. The framework for the accent recognition task begins with framing the signal the computing features, removing the silences using VAD. The accent recognition further involves training the Universal Background Model(UBM), computing statistics and classifying using ANN. The speaker recognition task uses the same framework till VAD as the former problem. It further involves computing bottle- neck features using a specially trained neural network for an utterance and computing the cosine distance to identify the speaker.
If you are interested in knowing more about this work, please ping me at [email protected]