Adaptive-order fractional Fourier transform features for speech recognition

We propose an acoustic feature for speech recognition based on the combination of MFCC and fractional Fourier transform (FrFT). Since the transform order is critical for the performance of FrFT, we use the ambiguity function to adaptively determine the optimal orders of FrFT for each frame. The performance of the proposed feature is compared with traditional MFCCs on recognizing speech of isolated and connected digits under both clean and noisy backgrounds. The recognition results and detailed confusion matrices are given and analyzed, which implies that the proposed feature is promising in certain speech processing fields.