peterf@0: #!/usr/bin/python
peterf@0: #
peterf@0: # annotationkit_create_annotation_protocol.py:
peterf@1: #    Prepare annotation protocol CSV from list of audio files
peterf@1: #
peterf@1: #    Read list of files from standard input and write CSV to standard output
peterf@1: #    See annotationkit_create_annotation_protocol_wrapper.sh for usage example
peterf@0: #
peterf@0: # Author: Peter Foster
peterf@1: # (c) 2015 Peter Foster
peterf@0: #
peterf@0: 
peterf@0: import fileinput
peterf@0: from scikits.audiolab import Sndfile
peterf@0: from pandas import DataFrame
peterf@0: import numpy as np
peterf@0: import sys
peterf@0: 
peterf@0: #Maximum number of chunks to sample from each file
peterf@0: nChunksPerFile = np.inf
peterf@0: #Duration of each chunk in seconds
peterf@0: chunkDuration = 4
peterf@0: #Expected sample rate
peterf@0: sampleRate = 48000
peterf@0: 
peterf@0: AudioChunks = []
peterf@0: np.random.seed(4756)
peterf@0: 
peterf@0: for audioFile in fileinput.input():
peterf@0:     audioFile = audioFile.strip()
peterf@0:     sf = Sndfile(audioFile, "r")
peterf@0:     if sf.samplerate != sampleRate: raise ValueError("wanted sample rate %g - got %g." % (sampleRate, sf.samplerate))
peterf@0:     
peterf@0:     nChunksInFile = int(sf.nframes / (sf.samplerate * chunkDuration))    
peterf@0:     #Sample without replacement random chunks from file
peterf@0:     sampledChunks = np.random.choice(nChunksInFile, min(nChunksInFile,nChunksPerFile), replace=False)
peterf@0:     
peterf@0:     for chunk in sampledChunks:
peterf@0:         frameStart = chunk * chunkDuration * sf.samplerate
peterf@0:         AudioChunks.append((audioFile, chunk, frameStart))
peterf@0:         
peterf@0:     sf.close()
peterf@0:     
peterf@0:     
peterf@0: #Create DataFrame
peterf@0: AudioChunks = DataFrame(AudioChunks)
peterf@0: AudioChunks.columns = ['audiofile', 'chunk', 'framestart']
peterf@0: 
peterf@0: sys.stderr.write("Processed " + str(len(AudioChunks)) + " chunks in total, corresponding to " + str(len(AudioChunks) * chunkDuration / float(60)) + " minutes of audio.\n")
peterf@0: 
peterf@0: #Write to CSV (stdout)
peterf@0: AudioChunks.to_csv(sys.stdout)