Safe is a system for scalable and efficient audio feature extraction. Safe is primarily useful for running bulk extraction on large sets of audio data. The Performance page shows how Safe compares with other audio feature extraction libraries. For a more detailed description of the system and the motivation behind it, see the Scalable Audio Feature Extraction thesis paper.
Safe is divided into two projects:
- safe-core - The core library, feature extraction algorithms, and a command-line interface.
- safe-dist - Provides cluster capabilities and a web-interface for feature extraction.
To build and run Safe, you will need to have SBT installed. The following command will compile Safe and generate scripts for running the system:
sbt start-script
The simplest way to run Safe is through the command-line interface. Safe provides the following command-line options:
-i <file> | --input <file>
input path (file or directory) for audio file(s) to process
-r | --recursive
flag to process audio data in sub-directories (default = false)
-p <file> | --plan <file>
path to feature extraction plan file
-f <value> | --feature <value>
description and parameters for a single feature to extract
-s <value> | --sample-rate <value>
sample rate in Hz of all audio files (default = 44100)
-o <file> | --output-dir <file>
directory to write featre output to (default = './')
-m | --metrics
flag to capture/print metrics (default = false)
In order to run the system, an input audio file (or directory containing audio files) -i and a description of the feature(s) (-f or -p) to be extracted are required. Features are described in the following format (where <parameters> are a comma-separated list of param=value options):
name: Feature [ <parameters> ] [ -out <parameters> ]
name is a unique identifier that will be used in the name of the output file. Feature is the system identifier for the feature to be extracted. The following are some example descriptions of features:
my_mfcc: MFCC frameSize=512, stepSize=256, windowType="hamming"
my_flux: SpectralFlux
my_shape: SpectralShape -out precision=2
The following example will extract the Mel-Frequency Cepstral Coefficients from an audio file "/home/safe/music/song.wav" and write the results to "/home/safe/out/song.wav.mfcc.csv":
safe-core/target/run -i "/home/safe/music/song.wav" -f "mfcc: MFCC" -o "/home/safe/out"
See Existing Features for a full description of the current feature extraction algorithms and parameters.
Safe currently only works with PCM-encoded audio files (WAV, AIFF, etc.). Future versions will support more encodings such as MP3, OGG, etc.
This software is licensed under the Eclipse Public License 1.0 (http://www.eclipse.org/org/documents/epl-v10.html)
Yourkit - Kindly supporting open source projects with its full-featured JVM Profiler