SBIR-STTR Award

This Small Business Innovation Research Phase I project will research sound-object recognition algorithms for use by professional and consumer audio recording and live sound engineers. Algorithms for robust off-line instrument recognition, music loop retrieval, dialog/sound effect/music recognition, and on-the-fly machine listening will also be developed. Musicians and audio engineers have access to gigabytes of audio content yet, the state of the art for finding audio content is through text queries and navigating static file hierarchies. Currently, none of the audio software manufacturers provide tools for searching for audio loops by their audio content. Additionally, recording and live sound engineers have complex organization and navigation duties, which could be solved using real-time audio analysis algorithms. If successful, this effort will enable recognizing audio content using a top-down approach - using a fleet of hierarchical machine learning classifiers, trained on statistical features extracted from one of the largest real-world audio content collections. Developed off-line machine classifiers will be ported to real-time time, embedded machine-listening algorithms, and used to enhance traditional audio signal processing tools. Further, the effort will foster interaction and collaboration between industry and academia ? encouraging sponsored research agreements, guest lecturers from industry engineers, and courses which directly focus on solving applied, industry challenges. This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5)

This Small Business Innovation Research (SBIR) Phase II project includes research and development of audio recognition and analysis software for offline and real-time sound recognition. Musicians and audio engineers have access to terabytes of music loops and sound effects. However, musicians are limited to searching for sounds using text-only keyword searches. This is a mundane, inaccurate, and exhausting process that ignores the files' actual audio content. The proposed solution provides a unique "find-something-that-sounds-like-this" search engine. Media production software and hardware is too complex, tedious, and labor-intensive for both novice and advanced users. The proposed sound platform adds capability that was previously missing - recognizing an input sound and automatically choosing the best parameters for the user. This project uses a signal processing and machine learning platform to perform novel experiments for classifying audio streams in real-time, improving recognition accuracy, and retrieving sounds from large collections. Commercial-quality software development kits for offline and real-time sound recognition will be developed. This project integrates state-of-the-art machine learning, digital signal processing, and information retrieval techniques. If successful, the platform will be able to listen to an audio signal and understand what it is listening to - as human listeners can identify and classify sounds. This innovative technology will be licensed to audio and music technology software and hardware manufacturers. The platform is suited for long-term discoveries and innovation, with demonstrated commercial interest from biomedical signal processing, security/surveillance, and interactive gaming companies. In the first chosen market, (sound engineering) the platform will have direct cultural benefits for musicians, music hobbyists, and audio engineers. It will allow music creation and audio production to become a completely creative task - minimizing the tedious technical issues that hinder the creative process, and lowering the barriers to entry for novice musicians and creative professionals