The Impact of Room Acoustics on Replay Speech Signal

An audio recording is affected by the possible distortions and artifacts present in the acoustic environment, the quality of recording, etc. The sound signal experiences multiple reflections and transmissions from various surfaces in a room and hence, causes temporal and spectral smearing of the recorded sound and this distortion is referred to as reverberation. The replay spoof speech signals are generated by using the playback of the recorded speech signal. This replay signal involve double convolution operation, which results in smoothing (averaging) the signal. We observe the spectral energy density of speech signal using traditional spectrogram and with Teager energy-based approach. The replay signal generated in simulated (controlled) acoustic environment has high spectral energies across entire frequency regions compared to the real (uncontrolled) replay signal. We performed the experiments on ASVspoof 2019 LA and PA task using GMM and One Class (OC-softmax) classifiers with Teager Energy Cepstral Coefficients (TECC) resulting in Equal Error Rate (% EER) of 7.51 % and 4.32 % on evaluation set, respectively.