earticle

영어

Speech enhancement is the task of improving the quality of the speech by reducing the noise. The magnitude of the short-time Fourier transform(STFT) or spectrogram is widely used for speech enhancement. However, this approach neglects the noisy phase and limits the quality of enhancement. Recently, short-time discrete cosine transform(STDCT) has been introduced to overcome the limitation of the STFT. STDCT is a real value representation; thus, it does not require phase information to reconstruct the audio. This paper compares the two approaches and analyzes the importance of phase information in speech enhancement. Our experiment shows that when trained under similar condition STFT performs better than STDCT in low noise scenarios, however, for high noise situations, STDCT has better performance than STFT.