def main(input_filename, output_filename, tatums, beats, bars):
audiofile = audio.LocalAudioFile(input_filename)
num_channels = audiofile.numChannels
sample_rate = audiofile.sampleRate
# mono files have a shape of (len,)
out_shape = list(audiofile.data.shape)
out_shape[0] = len(audiofile)
out = audio.AudioData(shape=out_shape, sampleRate=sample_rate,numChannels=num_channels)
# same hack to change shape: we want blip_files[0] as a short, silent blip
null_shape = list(audiofile.data.shape)
null_shape[0] = 2
null_audio = audio.AudioData(shape=null_shape)
null_audio.endindex = len(null_audio)
low_blip = audio.AudioData(blip_filenames[0])
med_blip = audio.AudioData(blip_filenames[1])
high_blip = audio.AudioData(blip_filenames[2])
all_tatums = audiofile.analysis.tatums
all_beats = audiofile.analysis.beats
all_bars = audiofile.analysis.bars
if not all_tatums:
print "Didn't find any tatums in this analysis!"
print "No output."
sys.exit(-1)
print "going to add blips..."
for tatum in all_tatums:
mix_list = [audiofile[tatum], null_audio, null_audio, null_audio]
if tatums:
print "match! tatum start time:" + str(tatum.start)
mix_list[1] = low_blip
if beats:
for beat in all_beats:
if beat.start == tatum.start:
print "match! beat start time: " + str(beat.start)
mix_list[2] = med_blip
break
if bars:
for bar in all_bars:
if bar.start == tatum.start:
print "match! bar start time: " + str(bar.start)
mix_list[3] = high_blip
break
out_data = audio.megamix(mix_list)
out.append(out_data)
del(out_data)
print "blips added, going to encode", output_filename, "..."
out.encode(output_filename)
print "Finito, Benito!"
def mashComponents(localAudioFiles, loudnessMarkers):
instSegments = localAudioFiles[0].analysis.segments# This is the base track
vocalSegments = localAudioFiles[1].analysis.segments# This is the overlay track
instBeats = localAudioFiles[0].analysis.beats[loudnessMarkers[0][0]:
loudnessMarkers[0][1]]
vocalBeats = localAudioFiles[1].analysis.beats[loudnessMarkers[1][0]:
loudnessMarkers[1][1]]
pitches = meanPitches(instSegments,instBeats)
timbre = meanTimbre(instSegments,instBeats)
sections = localAudioFiles[1].analysis.sections #This is the new lead vocal layer
sections = sections.that(selection.are_contained_by_range(
vocalBeats[0].start, vocalBeats[-1].start+vocalBeats[-1].duration))
if(len(sections)==0):sections = localAudioFiles[1].analysis.sections[2:-2]
pyplot.figure(0,(16,9))
image = numpy.array(pitches)
image = numpy.concatenate((image,numpy.array(timbre)),axis = 1)
image = numpy.concatenate((image,numpy.array(meanLoudness(instSegments,instBeats))),
axis = 1)
""" Now image contains chromatic, timbral, and loudness information"""
sectBeats = getSectBeats(sections[0]) # get beats that comprise a specific section
template = numpy.array(meanPitches(vocalSegments,sectBeats))
template = numpy.concatenate((template,numpy.array(
meanTimbre(vocalSegments,sectBeats))),axis=1)
template = numpy.concatenate((template,numpy.array(
meanLoudness(vocalSegments,sectBeats))),axis = 1)
im = feature.match_template(image,template,pad_input=True)
maxValues = [] #tuples of x coord, y coord, correlation, and section len(in secs)
ij = numpy.unravel_index(numpy.argmax(im), im.shape)
x, y = ij[::-1]
maxValues.append((numpy.argmax(im),x,y,sections[0].duration))
for i in range(len(sections)-1):
sectBeats = getSectBeats(sections[i+1])
template = numpy.array(meanPitches(vocalSegments,sectBeats))
template = numpy.concatenate((template,numpy.array(
meanTimbre(vocalSegments,sectBeats))), axis=1)
template = numpy.concatenate((template,numpy.array(
meanLoudness(vocalSegments,sectBeats))),axis = 1)
match = feature.match_template(image,template,pad_input=True)
ij = numpy.unravel_index(numpy.argmax(match), match.shape)
x, y = ij[::-1]
maxValues.append((numpy.argmax(match),
TEMPLATE_WIDTH*i+x,y,sections[i+1].duration))
im = numpy.concatenate((im,match),axis = 1)
maxValues.sort()
maxValues.reverse()
try:
count = 0
while(maxValues[count][3] < 10.0): # choose a section longer than 10 secs
count += 1
x = maxValues[count][1]
y = maxValues[count][2]
except:
print "exception in mashComponents..."
ij = numpy.unravel_index(numpy.argmax(im), im.shape)
x, y = ij[::-1]
pyplot.imshow(im, cmap = pyplot.get_cmap('gray'), aspect = 'auto')
pyplot.plot(x,y,'o',markeredgecolor='r',markerfacecolor='none',markersize=15)
pyplot.show()
sectionBeats = getSectBeats(sections[x/TEMPLATE_WIDTH])
print "len(sectionBeats): ", len(sectionBeats)
print "len(instBeats): ", len(instBeats)
print "y: ", y
y = instBeats[y].absolute_context()[0]
instBeats = localAudioFiles[0].analysis.beats
matchingBeats = instBeats[(y-len(sectionBeats)/2):(y+len(sectionBeats)/2)]
print"len(matchingBeats): ", len(matchingBeats)
matchingBeats = matchingBeats[-len(sectionBeats):]
print"len(matchingBeats): ", len(matchingBeats)
""" Check to make sure lengths of beat lists are equal... """
if len(matchingBeats) != len(sectionBeats):
print "len(matchingBeats) != len(sectionBeats). For now, I will just truncate..."
print "len(matchingBeats): ", len(matchingBeats)
print "len(sectionBeats): ", len(sectionBeats)
if len(matchingBeats) > len(sectionBeats):matchingBeats = matchingBeats[
:len(sectionBeats)]
else: sectionBeats = sectionBeats[:len(matchingBeats)]
""" I have to make sure sectionBeats and matchingBeats are similarly aligned
within their group, aka bar of four beats. I will add a beat to the beginning
of matchingBeats until that condition is met. I re-initialize instBeats and
vocalBeats, because now I want to include the areas outside of those marked
off by AutomaticDJ for fade ins and fade outs."""
vocalBeats = localAudioFiles[1].analysis.beats
while(matchingBeats[0].local_context()[0] != sectionBeats[0].local_context()[0]):
matchingBeats.insert(0,instBeats[matchingBeats[0].absolute_context()[0]-1])
sectionBeats.append(vocalBeats[sectionBeats[-1].absolute_context()[0]+1])
""" Check to make sure lengths of beat lists are equal... """
if len(matchingBeats) != len(sectionBeats):
print "len(matchingBeats) != len(sectionBeats) at the second checkpoint."
print "This should not be the case. The while loop must not be adding beats"
print "to both lists equally."
print "len(matchingBeats): ", len(matchingBeats)
print "len(sectionBeats): ", len(sectionBeats)
sys.exit()
""" Next, I will use the beats around the designated beats above to transition into
and out of the mashup. """
XLEN = 4 # number of beats in crossmatch
if(matchingBeats[0].absolute_context()[0] < XLEN or
len(instBeats) - matchingBeats[-1].absolute_context()[0] - 1 < XLEN or
sectionBeats[0].absolute_context()[0] < XLEN or
len(vocalBeats) - sectionBeats[-1].absolute_context()[0] - 1 < XLEN):
XLEN -= 1
BUFFERLEN = 12 # number of beats before and after crossmatches
while(matchingBeats[0].absolute_context()[0] < BUFFERLEN+XLEN or
len(instBeats) - matchingBeats[-1].absolute_context()[0] - 1 < BUFFERLEN+XLEN or
sectionBeats[0].absolute_context()[0] < BUFFERLEN+XLEN or
len(vocalBeats) - sectionBeats[-1].absolute_context()[0] - 1 < BUFFERLEN+XLEN):
BUFFERLEN -= 1
try:
""" These are the 4 beats before matchingBeats. These are the four beats of the
instrumental track that preclude the mashed section. """
b4beatsI = instBeats[matchingBeats[0].absolute_context()[0]-XLEN:
matchingBeats[0].absolute_context()[0]]
""" These are the 4 beats after matchingBeats. These are the four beats of the
instrumental track that follow the mashed section. """
afterbeatsI = instBeats[matchingBeats[-1].absolute_context()[0]+1:
matchingBeats[-1].absolute_context()[0]+1+XLEN]
if(len(b4beatsI) != len(afterbeatsI)):
print "The lengths of b4beatsI and afterbeatsI are not equal."
""" These are the 16 beats before the 4-beat crossmatch into matchingBeats. """
preBufferBeats = instBeats[matchingBeats[0].absolute_context()[0]-BUFFERLEN-XLEN:
matchingBeats[0].absolute_context()[0]-XLEN]
""" These are the 16 beats before the 4-beat crossmatch into matchingBeats. """
postBufferBeats = instBeats[matchingBeats[-1].absolute_context()[0]+1+XLEN:
matchingBeats[-1].absolute_context()[0]+1+XLEN+BUFFERLEN]
if(len(preBufferBeats) != len(postBufferBeats)):
print "The lengths of preBufferBeats and postBufferBeats are not equal."
print "len(preBufferBeats): ", len(preBufferBeats)
print "len(postBufferBeats): ", len(postBufferBeats)
print matchingBeats[-1].absolute_context()[0]
print len(instBeats)
sys.exit()
""" These are the 4 beats before matchingBeats. These are the four beats of the
new vocal track that preclude the mashed section. """
b4beatsV = vocalBeats[sectionBeats[0].absolute_context()[0]-XLEN:
sectionBeats[0].absolute_context()[0]]
""" These are the 4 beats after matchingBeats. These are the four beats of the
new vocal track that follow the mashed section. """
afterbeatsV = vocalBeats[sectionBeats[-1].absolute_context()[0]+1:
sectionBeats[-1].absolute_context()[0]+1+XLEN]
if(len(b4beatsV) != len(afterbeatsV)):
print "The lengths of b4beatsI and afterbeatsI are not equal."
sys.exit()
except:
print "exception in 4 beat try block."
sys.exit()
""" vocData: An AudioData object for the new vocal data that will be overlaid.
instData: An AudioData object for the base instrumental track.
originalVocData: An AudioData object of the original vocal to accompany
the new one.
vocalMix: An AudioData of both vocal tracks mixed together, in order to
keep the overall vocal loudness approximately constant.
mix: An AudioData of the instrumental track and combined vocals
mixed together. """
vocData = audio.getpieces(localAudioFiles[3],b4beatsV+sectionBeats+afterbeatsV)
instData = audio.getpieces(localAudioFiles[2],b4beatsI+matchingBeats+afterbeatsI)
if instData.data.shape[0] >= vocData.data.shape[0]:
mix = audio.megamix([instData, vocData])
else:
mix = audio.megamix([vocData, instData]) # the longer data set has to go first.
mix.encode('mix.mp3')
vocData.encode('vocData.mp3')
""" Now, make a similar mix for before the mashed sections..."""
instData = audio.getpieces(localAudioFiles[2], preBufferBeats + b4beatsI)
vocData = audio.getpieces(localAudioFiles[4], preBufferBeats + b4beatsI)
premix = audio.megamix([instData, vocData])
""" ...and another mix for after the mashed sections."""
instData = audio.getpieces(localAudioFiles[2], afterbeatsI + postBufferBeats)
vocData = audio.getpieces(localAudioFiles[4], afterbeatsI + postBufferBeats)
postmix = audio.megamix([instData, vocData])
""" Now, I have three AudioData objects, mix, premix, and postmix, that overlap by
four beats. I will build Crossmatch objects from the overlapping regions, and three
Playback objects for the areas that are not in transition. """
action.make_stereo(premix)
action.make_stereo(mix)
action.make_stereo(postmix)
preBuffdur = sum([p.duration for p in preBufferBeats]) # duration of preBufferBeats
playback1 = action.Playback(premix,0.0,preBuffdur)
b4dur = sum([p.duration for p in b4beatsI]) # duration of b4beatsI
crossfade1 = action.Crossfade((premix,mix),(preBuffdur,0.0),b4dur)
abdur = sum([p.duration for p in afterbeatsI])
playback2 = action.Playback(mix,b4dur,mix.duration - b4dur - abdur)
crossfade2 = action.Crossfade((mix,postmix),(mix.duration - abdur,0.0),abdur)
playback3 = action.Playback(postmix,abdur,sum([p.duration for p in postBufferBeats]))
action.render([playback1,crossfade1,playback2,crossfade2,playback3], 'mashup.mp3')
# we need to add silence, because segment1 is longer if num_channels > 1: silence_shape = (reference_data.endindex,num_channels) else: silence_shape = (reference_data.endindex,) new_segment = audio.AudioData(shape=silence_shape, sampleRate=out.sampleRate, numChannels=segment_data.numChannels) new_segment.append(segment_data) new_segment.endindex = len(new_segment) segment_data = new_segment elif reference_data.endindex < segment_data.endindex: # we need to cut segment2 shorter, because segment2 is shorter index = slice(0, int(reference_data.endindex), 1) segment_data = audio.AudioData(None, segment_data.data[index], sampleRate=segment_data.sampleRate) else: # TODO: stretch samples to fit. # haven't written this part yet. segment_data = segment_data segment_data_list.append(segment_data) # mix the original and the remix #mixed_data = audio.mix(segment_data,reference_data,mix=mix) mixed_data = audio.megamix(segment_data_list) out.append(mixed_data) # redner output out.encode(outputFilename)
else:
silence_shape = (reference_data.endindex, )
new_segment = audio.AudioData(
shape=silence_shape,
sampleRate=out.sampleRate,
numChannels=segment_data.numChannels)
new_segment.append(segment_data)
new_segment.endindex = len(new_segment)
segment_data = new_segment
elif reference_data.endindex < segment_data.endindex:
# we need to cut segment2 shorter, because segment2 is shorter
index = slice(0, int(reference_data.endindex), 1)
segment_data = audio.AudioData(
None,
segment_data.data[index],
sampleRate=segment_data.sampleRate)
else:
# TODO: stretch samples to fit.
# haven't written this part yet.
segment_data = segment_data
segment_data_list.append(segment_data)
# mix the original and the remix
#mixed_data = audio.mix(segment_data,reference_data,mix=mix)
mixed_data = audio.megamix(segment_data_list)
out.append(mixed_data)
# redner output
out.encode(outputFilename)