This paper proposes an algorithm to perform monaural speech source separation by means of time-frequency masking. The algorithm is based on the estimation of the power spectrum of the original speech signals as a combination of a carrier signal multiplied by an envelope. A Multi-Frequency Harmonic Product Spectrum (MF-HPS) algorithm is used to estimate the fundamental frequency of the signals in the mixture. These frequencies are used to estimate both the carrier and the envelope from the mixture. Binary masks are generated comparing the estimated spectra of the signals. Results show an important improvement in the separation in comparison to the original algorithm that only uses the information from the HPS.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=16733
(352KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
