Blind One Microphone Speech Separation Based On Pitch Detection

Blind Source Separation (BSS) is a challenging research subject and becomes to be a popular research area in signal processing field in recent years. In the last two decades, a body of algorithms addressing the instantaneous BSS problems have been proposed and gained some results. Various applications can be found in speech processing, array processing, biomedical signal processing, image processing and communications. However, current research results are far from its mature solution, especially in the problem of blind one microphone speech separation. Compared with the problem of multi-input signals, blind one microphone speech separation is much more difficult for less information to be used and the separation results are still not satisfactory. It is still an open research subject in speech signal processing.This thesis first took a look at the BSS research history and then concentrated on the characteristics and the different performance in the spectrogram of speech signals. A new BSS algorithm was then proposed based on these characteristics and differences. Specifically, we firstly compute the linear prediction residual of the speech signals, and the cepstrum of these residual are then utilized for pitch determination. Based on the spectrogram of the mixed signals, these speech pitches combining with several properties of speech signals (e.g., energy, continuity of speech pitches, and zero-crossing rate of short-time segments) are leveraged to identify the target speech signal, while the non-target signal is filtered out from the spectrogram using spectrum subtraction method. Finally, the target speech signal is restored from the remaining spectrogram. We conducted extensive experiments to evaluate the proposed method. The experimental results indicate that our method can effectively separate the speech signals of different speakers under the environment of anechoic or poor echo. Compared to existing method, our approach is superior in terms of performance and efficiency.