Time-Scale Modification Phase Vocoder for Accent Recognition

English is an international language for most countries in this world use to communicate. Although there are only common English versions, namely UK and US, there are differences in pronunciation due to the influence of the first language. An accent is a special trait in human communication, which cannot be obtained through education but is obtained when someone lives in a unique and cultured area for a long time. Accent recognition is used in both forensic applications and Automatic Speech Recognition Systems (ASR) but accent variations are required for speaker recognition or speech recognition, therefore accent recognition research still has many challenges. In this research, phase vocoder is used as a time-scale modification method used to equate all audio data to the same length duration to the average duration of each first language. This research and related researches that use the same dataset employ imbalanced data for analysis. Then after augmentation of audio data, spectral-based features are used and Linear SVM as a classifier with hyperparameter regularization. From the experiment has been done performance classification the lowest is 45% and the highest is 84.1%.