Music and Audio Processing

Problem statement

More people enjoy music from high-capacity portable players and from internet music services. Developing natural methods for exploring the growing music collections is now a critical need. In this work we developed a prototype system for interactive, computer-assisted music browsing based on the direct presentation of short music thumbnails.

Music Genie refers to audio processing technologies for quick and easy music browsing and discovery. We built research prototype software that allows users to quickly generate playlists from a collection of thousands of songs. Our software provides an interactive, content-adaptive, thumbnail-based presentation of music. The software accepts user inputs interactively during use, and it consecutively presents short music thumbnails for quick browsing. This direct, interactive presentation of music makes our system simple and transparent for users. Unobstrusively, the user's interactions (or lack thereof) are input into a real time system that, based on the prior analysis of the music content, adaptively presents a potentially interesting new song thumbnail. This technology may be used to 1) assist users discover music for purchase from a web music store; 2) allow users to browse their PC music collection based on current taste; and 3) allow intelligent browsing, playback or playlist generation in small, interface-limited, music players.
 

A white paper describing the technology is found here.

Last Modified: Monday, July 23, 2007 11:01:40 -0700

Reducing Audio Noise Using Spectrogram Random Textures

Many are unpleasantly surprised by the unexpected noise and the low quality of the audio of their home video clips. The goal here is to improve the quality of consumer generated audio: audio captured using consumer electronics devices under uncontrolled conditions. Making the problem harder, consumer-generated audio is not speech alone. Many indoor audios contain music, and many outdoor audios contain important natural sounds such as waterfalls, ocean waves, etc. In this work, we developed techniques for enhancing consumer-generated audio. The example described below applies to digital camera zoom motor noise, a degradation commonly found in short video clips.

Outline

his work involves audio enhancement when a strong, additive noise is present only during a known or easily detected period of moderate length (of around one second). The signals may contain intelligible components such as speech or music, and may also contain desired, but unintelligible, background components such as rivers or waterfalls. A first estimate synthesizes the unintelligible components from the noise-free neighboring spectrogram. A second estimate recovers the intelligible components using spectral attenuation. The two estimates are combined using ideas from statistical process control. Tests with audio containing digital camera zoom motor noise, and with simulations, validate the approach.

Technical details are found here.

Last Modified: Wednesday, 07/18/2007 13:55:47 -0700

For further information, contact: