This is the list of research projects conducted at Gracenote and Stanford. You may see the details by clicking their names.

Multiple Pitch Perception Based on Neural Interspike Interval Histogram

Recognizing multiple pitches of musical instruments is still a hard problem in music information retrieval. We describe a model for multiple pitch perception inspired by the human auditory system. Example sample sounds used for this project can be downloaded from here.

Techniques: Sparse representation, Interspike interval histogram, Audio DSP

Music vs. Noise Classification

In this paper a novel audio classification method based on Entropy Model is presented. By applying this model on Harmonic Product Spectrum, a promising classification power between music and various ambient noises can be achieved.

Techniques: Entropy, Audio DSP

Fast Philips Fingerprinting Algorithm with Minimal Artifacts

In this paper we introduce a modified Philips fingerprinting algorithm which generates fingerprints twice faster than the original Philips algorithm. Our test results show promising music recognition rates from this new algorithm.

Techniques: Audio DSP

Humming Control Interface for Hand-held Devices

Sook Young Won, Dong-In Lee, Julius Smith, Proceedings of International ACM SIGACESS Conference on Computers and Accessibility, Tempe, Arizona, October 2007

Techniques: HCI, Pitch detection algorithm

Comparison of Normal and Husky Voice using Self Organizing Map

Self-organizing map (SOM) is useful for visualizing low-dimensional views of high-dimensional data. This project aims at visualizing the map trained by voice signals. Whereas the map obtained from normal voice shows regular and steady patterns, there is irregularity in the map generated from husky voice. The result suggests that incidental exposures to husky voice might subsequently degrade the listener's normal musical ability.

Techniques: Self-organizing map, Pitch class profiling

Query by Humming System

Query by humming (QbH) is a music retrieval system that takes a user-hummed melody, and compares it to an existing database. The system then returns a ranked list of music closest to the input query. In this project, my program gets the hummed melody through a microphone, and display the closest music in the test database.

Techniques: Dynamic programming, Pitch detection algorithm


"Interactive Streaming Content Apparatus, Systems, and Methods"

U.S.A., Michael Jeffrey, Markus K. Cremer, Dong-In Lee, Gracenote, Inc.,
Application #- 61/502,799, June 29 2011