Speech Source Separation Using a Multi-Pitch Harmonic Product Spectrum-Based Algorithm
This paper presents an efficient algorithm for separating speech signals by determining multiple pitches from mixtures of signals and assigning the sources to one of those estimated pitches. The pitch detection algorithm is based on Harmonic Product Spectrum. Since the pitch of speech signals fluctuates readily, a frame-based algorithm is used to extract the multiple pitches in each frame. Then, the fundamental frequency (pitch) for each source is estimated and tracked after comparing all the frames. The estimated fundamental frequency of the sources is then used to generate a set of binary masks that allow separating the signals in the Short Time Fourier Transform domain. Results show a considerable separation of the speech signals, justifying the feasibility of the proposed method.
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 and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is temporarily free for AES members.