I am looking to develop an Android app. As part of the functionality the app would require to randomly sample 3-5 seconds of audio and classify it as containing human speech or not. I understand that this concept is called Voice Activity Detection?
What would be the best way to implement this on a Mobile phone. I developed a basic system using energy based features and thresholds. I am hoping to find something less susceptible to noise, probably using features such as MFCC or formants? I did go through a number of papers, but most of them would require me to collect data and train models. Is there any library or framework I could use which would work in realtime?