I was talking to a friend about the spectral analysis of EEG signals. I was explaining to him how spectral analysis works and why it is so important. To explain him well, I used an analogy of speech processing and recognition. Going further, an idea crossed my mind about how we could use speech recognition to create written lyrics by doing spectral analysis of songs. So I thought about it as a great idea.
Haven’t I explained it well? So to elaborate it further, a lyrical transcription would take as input a song. It would do spectral analysis of it to extract the sound of the vocalist. It would understand what has been sung. And for what has been sung it would produce lyrics and show it on the screen with the song in real-time. As simple as that! But the system itself would not be as simplistic as its succinct description.
But the system itself would not be as simplistic as its succinct description. You will have to create a spectral analyzer. You will have to feed the spectral coefficients to a time-series prediction system such as a hidden Markov Model (HMM). You will have to train the HMM before that. And you will have to maintain a handful of databases and corpus to do all of this. However, it is interesting albeit challenging. It could be quite rewarding nonetheless. Remember, speech recognition is considered as an insurmountable opportunity. Following resources could be really good for getting this work done.
Matlab code and usage examples for RASTA, PLP, and MFCC speech recognition feature calculation routines, also inverting features to sound.
A tutorial on hidden Markov models and selected applications in speech recognition – IEEE Xplore Document
This tutorial provides an overview of the basic theory of hidden Markov models (HMMs) as originated by L.E. Baum and T. Petrie (1966) and gives practical d
HMM using Kevin Murphy matlab toolbox. Learn more about hmm
If you found an error, highlight it and press Shift + Enter or click here to inform us.