Browse Prior Art Database

Search/recognition algorithm for sound files.

IP.com Disclosure Number: IPCOM000100062D
Original Publication Date: 2005-Mar-15
Included in the Prior Art Database: 2005-Mar-15
Document File: 4 page(s) / 47K

Publishing Venue

IBM

Abstract

Searching algorithms (internet, filesystem, etc.) for documents have reached a plateau of maturity when the content being sought is in the form of textual documents, or if the object has been associated with certain keywords or other searchable metadata. The key is that the searching, caching, indexing, etc. is all based on textual data. Searching for binary data (assuming no textual metadata exists) is a much more nascent art. Some progress has been made in the realm of image searching, but this article focuses on searching for sound files - specifically music files.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 42% of the total text.

Page 1 of 4

Search/recognition algorithm for sound files .

Do you ever hear a song on the radio that you really like, but you can't find out what it's called or who it's by? You go around for days with one refrain bouncing around in your head. Maybe you didn't even catch any of the lyrics, but you've got the basic melody of the chorus. Proposed is an algorithm and system whereby you could search for that song based on a snippet you can 'sing' or otherwise perform audibly into your computer's microphone.

Searching algorithms (internet, filesystem, etc.) for documents have reached a plateau of maturity when the content being sought is in the form of textual documents, or if the object has been associated with certain keywords or other searchable metadata. The key is that the searching, caching, indexing, etc. is all based on textual data. Searching for binary data (assuming no textual metadata exists) is a much more nascent art. Some progress has been made in the realm of image searching, but this disclosure focusses on searching for sound files - specifically music files.

Do you ever hear a song on the radio that you really like, but you can't find out what it's called or who it's by? You go around for days with one refrain bouncing around in your head. Maybe you didn't even catch any of the lyrics, but you've got the basic melody of the chorus. This invention proposes an algorithm and system whereby you could search for that song based on a snippet you can 'sing' or otherwise perform audibly into your computer's microphone.

This invention consists of:

(1) An Analysis and Indexing Algorithm.

(2) An Indexing Database.

(3) A Search Algorithm.

(4) An Application Programming Interface (API) to the above.

A front-end application can use the above components (via the API) along with existing technology. A 'setup' application would do the following:
a) Analyze and index a library of sound files via the Analysis and Indexing Algorithm (1)
b) Store the resulting data along with the desired identifiers (e.g. song title and artist) in the Indexing Database (2)

The 'user' front end would do the following:
a) Procure the audio sample to be searched for ("Search Sample") via, for example, existing sound recording technology;
b) Provide configurable parameters ("Tunables", appropriately) for the search, e.g. via a web form;

1

Page 2 of 4

     c) Invoke the Analysis and Indexing Algorithm (1) to generate metadata on the Search Sample;
d) Pass the metadata and Tunables to the Search Algorithm (3);
e) Process and format the results for the user.

The core idea of this invention is based around the concept that a piece of music can be described partially in terms of specific notes, intervals between those notes in a melody line, and the duration of each note (a superset of the tempo of the piece). The invention consists of a system and algorithm for describing and classifying chunks of sound based on these criteria.

(1) Analysis and Indexing Algorithm.

Given a sound file, t...