I am experimenting with Multi-modal content recommendation for some TV programs. For now i am working in python on audio features to find out the programs similarities with each-other.
I want to input some audio features into a Bag-of-Audio words. Features are extracted from the complete audio of programs/episodes containing voice-overs as well.
The question is, if MFCC features are extracted directly on program's background music+voiceovers, do they produce any promising results?
which are the other audio features that may work well for audios when combined with voice-overs?
If not then what can be the possible method/software/technique to separate the audio and voice overs from a program episode? and whether thy work better when processed separately and then combined later on?
- I have tried using some software like Audacity to separate the music with voice-over and it extremely reduces the audio quality.