def auditory_spectrum(audio_file_path):
''' computes an auditory spectrogram based from an acoustic array '''
import dear.io as io
decoder = io.get_decoder(name='audioread')
audio = decoder.Audio(audio_file_path)
st = 0
graph = 'Y5'
N = 64
win = 0.025
hop = 0.010
freqs = [110., 2 * 4435.]
combine = False
spec = [[]]
gram = getattr(auditory, graph)
gram = gram(audio)
for t, freq in enumerate(
gram.walk(N=N,
freq_base=freqs[0],
freq_max=freqs[1],
start=st,
end=None,
combine=combine,
twin=win,
thop=hop)):
spec[0].append(freq)
return np.array(spec), audio._duration
assert output is not None
assert algorithm in CROP_ALGORITHMS
assert 0 < length
assert (end is None) or 0 <= start < end
assert 0 < samplerate
assert 0 < bitrate
except Exception as ex:
print_exc()
exit_with_usage()
if len(args) != 0:
exit_with_usage()
func = CROP_ALGORITHMS.get(algorithm)
import dear.io as io
decoder = io.get_decoder(name='audioread')
audio = decoder.Audio(inputf)
print "SampleRate: %d Hz\nChannel(s): %d\nDuration: %d sec"\
% (audio.samplerate, audio.channels, audio.duration)
if start >= audio.duration:
print "[error] Start time is beyond song duration. Not cropping."
else:
start, end = func(audio, start, end, length)
duration = end - start
print start, end, duration
tmpfile = output + '.crop.tmp.wav'
cmd = [FFMPEG,'-i',inputf,'-ss',str(start),'-t',str(duration),
'-ac','1','-ar',str(samplerate),'-ab',bitrate,
tmpfile]
[-h] step size, default 1024
[-f] frequency boundary, default (0, 7040)
"""
exit()
try:
opts, args = getopt.getopt(sys.argv[1:], "g:s:t:o:h:w:q:n:f:b:rc")
except getopt.GetoptError as ex:
print ex
exit_with_usage()
if len(args) != 1:
#print args
exit_with_usage()
import dear.io as io
decoder = io.get_decoder(name='audioread')
audio = decoder.Audio(args[0])
print "SampleRate: %d Hz\nChannel(s): %d\nDuration: %d sec"\
% (audio.samplerate, audio.channels, audio.duration)
graph = 'dft'
st = 0
to = None
outfile = None
norm=colo.LogNorm(vmin=0.000001)
for o, a in opts:
if o == '-s':
st = float(a)
elif o == '-t':
to = float(a)
assert algorithm in CROP_ALGORITHMS
assert 0 < length
assert (end is None) or 0 <= start < end
assert 0 < samplerate
assert 0 < bitrate
except Exception as ex:
print_exc()
exit_with_usage()
if len(args) != 0:
exit_with_usage()
func = CROP_ALGORITHMS.get(algorithm)
import dear.io as io
decoder = io.get_decoder(name="audioread")
audio = decoder.Audio(inputf)
print "SampleRate: %d Hz\nChannel(s): %d\nDuration: %d sec" % (audio.samplerate, audio.channels, audio.duration)
if start >= audio.duration:
print "[error] Start time is beyond song duration. Not cropping."
else:
start, end = func(audio, start, end, length)
duration = end - start
print start, end, duration
tmpfile = output + ".crop.tmp.wav"
cmd = [
FFMPEG,
"-i",
inputf,
def render_file(fin, outdir, shape = (512,512), framerate = 25, sym = 6, inv = 1, pad = True, mode = 'dft', preserve_alpha=False, params = {}, no_color=False, edge_filter=True):
decoder = io.get_decoder(name = 'audioread')
audio = decoder.Audio(fin)
if not os.path.isdir(outdir):
os.mkdir(outdir)
fs = audio.samplerate
nframes = audio.duration * framerate
print 'fs: %d Hz, Duration: %d sec, Frames: %d'%(audio.samplerate,audio.duration,nframes)
gram = None
log_index = True
n_octaves = None
gram_args = []
gram_kwargs = {'start':0,'end':None}
if mode == 'dft':
win = 2 * fs / framerate
hop = fs / framerate
nfft = None
if 'w' in params:
win = params['w']
if 'n' in params:
nfft = params['n']
gram = dft.PowerSpectrum(audio,nfft=nfft)
gram_args = [win,hop]
elif mode == 'cnt':
n = 60
hop = 1.0 / framerate
n_octaves = 8
log_index = False
if 'n' in params:
n = params['n']
if 'o' in params:
n_octaves = params['o']
gram_args=[n]
gram_kwargs['hop'] = hop
gram_kwargs['freq_base'] = cqt.A0 * 4
gram_kwargs['freq_max'] = cqt.A0 * 2**n_octaves
gram = cqt.CNTPowerSpectrum(audio)
elif mode == 'gmt':
n = 60
hop = 1.0 / framerate
n_octaves = 9
log_index = False
if 'n' in params:
n = params['n']
if 'o' in params:
n_octaves = params['o']
n = n * (n_octaves-1)
gram_args = [n]
gram_kwargs['thop'] = hop
gram_kwargs['twin'] = 2*hop
gram_kwargs['freq_base'] = cqt.A0 * 4
gram_kwargs['freq_max'] = cqt.A0 * 2**n_octaves
gram_kwargs['combine'] = True
gram = auditory.GammatoneSpectrum(audio)
i = 0
j = 0
#print gram_args
#print gram_kwargs
tqcv = 0
iso226_factors = None
timesums=np.zeros(4)
for spectrum in gram.walk(*gram_args,**gram_kwargs):
iso_init = isinstance(iso226_factors, np.ndarray)
i += 1
if hasattr(gram, 'fqs') and not iso_init:
iso226_factors = np.array(map(iso,gram.fqs))
iso226_factors = 10 ** (iso226_factors / 10)
iso226_factors = 1 / iso226_factors
if os.path.isfile(outdir+'conv%05d.png'%i):
continue
if iso_init:
spectrum = np.multiply(spectrum,iso226_factors)
j += 1
im,times = draw_spectrum(spectrum,shape,sym,inv,log_index,n_octaves,mode=mode,no_color=no_color,edge=edge_filter)
timesums += np.array(times)
imsave(outdir+'img%05d.png'%i,(im*255).astype(np.uint8)) # this is for saving the kernel images
t0=time.time()
im = convolve_quaternion(im, pad, preserve_alpha, no_color=no_color)
tqcv += time.time()-t0
imsave(outdir+'conv%05d.png'%i,im.astype(np.uint8))
if i%100==0:
ttot = tqcv + timesums.sum()
actual_nframes = nframes - (i - j)
pct = i / nframes
pct_diff = j / actual_nframes
eta = ttot / pct_diff - ttot
print '%d %4.4f%% %6.2fs elapsed %6.2fs eta %6.2f convolution %s timing' % (i, 100.0 * pct, ttot, eta, tqcv, str(timesums))
print 'done rendering',i,tqcv+timesums.sum(),tqcv,timesums
return