Approaches for Speech/Noise File Detection
This paper discuss about detecting speech files in a real world speech recognition task. Detecting files with small background speech or noise, changes the overall behaviour of the Interactive Voice Response System. Two different approaches to detect speech/noise files are proposed and explored further to determine which works better for detecting speech as opposed to detecting noise. Artificial neural network trained to recognize phones are used as an intermediate tool. Further, boundary conditions where neither approaches works are also discussed. Some popular publically available dataset, are used to validate the approaches.
Keywords - Noise Robustness, neural networks, Interactive Voice Response Systems