Method and Apparatus for Audio Source Separation

Audio source separation is an essential requirement for systems used in music production, voice assistants,  telecommunications, and more. The need arises because of complexities dealing with the overlapping or mixed sound sources in a given signal. Extracting individual sound sources from this tangled aggregate is critical for the clarity and quality of the output. Traditional approaches to audio separation often face difficulties, especially with live or continuous speech and music mixtures. These approaches usually struggle to separate sound sources in real time with individual source clarity. Consequently, they often yield inadequate results, lacking efficiency and accuracy, resulting from the intertwined nature of the composite audio signals and the difficulty in splitting frequencies precisely.

Technology Description

The invention revolves around co-channel audio source separation, a process that allows the isolation of different sources of sound from a composite signal. The process involves transforming a frequency-related representation of the acoustic signal into a compressed format. It uses pitch identifications within various regions of the signal to generate multiple pitch estimates over time. Another aspect of the invention includes processing a mixed acoustic signal to obtain various acoustic properties that are used to provide separate pitch estimates for each acoustic signal. This technology stands out because of its capability to effectively fragment an acoustic signal into individual frequency and time regions. This compressed, localized representation enables a more efficient estimation of pitches, facilitating the separation of complex acoustic signals. The ability of this technology to process mixed signals by grabbing different time-frequency regions distinguishes it from other traditional approaches that often struggle with the task of separating composite acoustic signals in real time.