Update: JAudioscout no longer being maintained. See Redis-AudioScout for a revised implementation as a Redis module. This makes it much easier to build and deploy.
AudioScout™ is an audio identification system that allows for reasonable distortion in the audio signal. The system consists of a server which responds to commands from client applications.
You can read more about it in this blog post: here
An API package for designing and building custom client applications can be found here JPHashAudio
You will also need libAudioData along with its java bindings for reading audio files.
This package consists of just the AudioScout server, auscoutd, and a CLI client application for testing purposes - called AudioScout.
Recall reaches a 95% accuracy. This is based on 110 queries against an index holding over 1200 audio tracks. The query clips represent a fairly serious telephone distortion with an added Gaussian noise. The test was conducted various clip durations and toggle settings. A toggle settings is number of bits to consider flipping when looking up candidate hash frames in the sequence.
To build the package,
mvn compile test package
This should put the executables in the target bin directory.
The server API is provided by the Auscoutd server. It provides the means for client applications to query and submit and otherwise interact with the audio data.
server modes:
- query - simple query by an audio signal. Server returns 0 or more possible matches.
- submit - submit new files to be indexed. Server returns integer id. (The integer id is need for later deletion).
- query - complex query by audio samples. Returns 0 or more possible matches, each with a ranking confidence score and integer position denoting its relative position in which it was matched to the candidate position.
- sync - sync command to synchronize new entries into the data store.
- threshold - change the global threshold for queries.
- delete - command to delete designated id values from system. Note: Use sparingly. Entries are not immediately deleted but are saved to a file for periodic sweeps of all entries. The time of day when this is performed is configurable.
- query - query with threshold. Same as mode 3 but with adjustable threshold. The other queries are use the global threshold at which the server is set.
The Auscoutd program can be accessed by the corresponding program scripts in the bin folder. Simply invoke with no arguments to get the usage information:
./Auscoutd
A log4j configuration file can be found at /etc/auscoutd.properties file. On execution, auscoutd.log file will be produced in the bin directory.
Two options must be provided: -homedir and -mdata, which are the locations of directories for file storage for the data stores. Be sure you have plentiful space in these directories.
There is an example client program, AudioScout, which can be accessed by the corresponding script in the bin directory. Simply invoke with no arguments for usage information:
./AudioScout
The 'print' option allows for printing out the audio hash values for each audio file in a directory.
MDataScroll is a utility to scroll through the indexed entries in the meta data index.