Vowel and Gender classification using multi-task deep learning architecture
The motivation of this coursework will be to explore the use of ML and DL techniques to predict on the Vocal dataset. This paper will present models for prediction of both gender and vowel classification of the audio files. Applications of these models and techniques could be used on single note analyses of the vocals and can be used to get the corresponding voice to text translation and can be particularly useful in improving existing systems particularly in songs to obtain the song lyrics where sounds vary in the duration and frequency spoken.
Modern audio classification uses deep learning techniques which reduces the requirement of musical knowledge which was previously required for designing good features. In many ways, the previous research methods that were used can help us better understand and speculate on the inner workings of some of the Deep Learning algorithm.
The Full Analysis of the different models can be found in the document below Full Report Analysis
A sample trained model has been provided for one of the models which you can load and run.
