def method_get_length(self):
result = pybass.BASS_ChannelGetLength(self.bass_handle,
pybass.BASS_POS_BYTE)
if result <= 0 and isinstance(self.stream, memory_stream):
result = self.stream.decode_length
return result
def mfeats_single(path, queue=None):
# uses mono data
# tic = time.time()
find_key = True
find_gain = False
key, bit, error = ('', 16, 0)
path = os.path.abspath(path)
mdx = makemdx(path)
# key_analysis_fs = 6000
# waveform_length = 2000
# waveform_oversampling = 20
# key_analysis_fs = 6000
# waveform_length = 2500
# waveform_oversampling = 20
waveform_length = 10000
waveform_oversampling = 20
best_highlight_duration_beat = 32 * 2
# best_highlight_duration_beat = 16 * 2
version = MFEATS_VERSION
init_bass_decode()
channel = get_channel(path)
hstream = pybass.BASS_StreamCreateFile(
False, path, 0, 0, pybass.BASS_STREAM_DECODE
| pybass.BASS_STREAM_PRESCAN | pybass.BASS_UNICODE)
fs = ctypes.c_float()
pybass.BASS_ChannelGetAttribute(hstream, pybass.BASS_ATTRIB_FREQ,
ctypes.byref(fs))
fs = int(fs.value)
hlength = pybass.BASS_ChannelGetLength(hstream, pybass.BASS_POS_BYTE)
duration = pybass.BASS_ChannelBytes2Seconds(hstream, hlength)
total_frame_length = hlength / 2
if hstream == 0:
error = 1
if error == 1:
mfeats_data = MFEATS(mdx=mdx,
path=path,
date=time.time(),
version=version,
error=1)
if queue is not None:
queue.put(mfeats_data)
return
else:
return mfeats_data
if total_frame_length / 2 < waveform_length:
waveform_length = total_frame_length / 2
frame_length = int(1.0 * total_frame_length / waveform_length) * 2
if int(frame_length) % 8 != 0:
frame_length += (8 - int(frame_length) % 8)
gap = total_frame_length / frame_length - waveform_length
waveform = numpy.linspace(0, 0, int(round(waveform_length + gap)))
highlight_raw_points = numpy.linspace(0, 0,
int(round(waveform_length + gap)))
frame_raw = numpy.arange(frame_length, dtype=ctypes.c_short)
jump = 1.0 * frame_length / waveform_oversampling
analyze_frame, tempo_frame, tempo_fs = ([], [], 200)
for cnt, frame_position in enumerate(
numpy.arange(0, total_frame_length - frame_length, frame_length)):
pybass.BASS_ChannelGetData(
hstream, frame_raw.ctypes.data_as(ctypes.POINTER(ctypes.c_short)),
int(frame_length * 2))
mono_frame = deepcopy(frame_raw[::channel])
analyze_frame += [mono_frame]
if jump < waveform_oversampling:
waveform[cnt] = numpy.max(numpy.abs(mono_frame))
else:
points = [
numpy.max(numpy.abs(mono_frame[int(i):int(i + jump)]))
for i in numpy.arange(0, frame_length / channel - jump, jump)
]
waveform[cnt] = numpy.mean(points)
highlight_raw_points[cnt] = numpy.mean(numpy.abs(mono_frame))
# collect frames for long term analysis
alength = len(analyze_frame) * len(analyze_frame[-1])
if alength >= fs * 30 or alength * channel >= total_frame_length - frame_length:
analyze_frame = numpy.concatenate(analyze_frame, axis=0)
num = int(len(analyze_frame) / (1.0 * fs / tempo_fs))
tempo_frame += [
numpy.abs(
numpy.interp(
numpy.linspace(0,
len(analyze_frame) - 1, num),
numpy.arange(len(analyze_frame)), analyze_frame))
]
# key_frame_length = int(fs*0.25); key_frame_jump = 0.8
# for i in range(0, len(analyze_frame)-key_frame_length,\
# int(key_frame_length*key_frame_jump)):
# spectrum = numpy.fft.fft(\
# analyze_frame[i:i+key_frame_length],\
# int(fs*key_analysis_resolution))
# spectrum = numpy.abs(spectrum[1:int(len(spectrum)/2)])
# notes = spectrum_to_note_by_max(spectrum, note_freq_div)
# chromagram += [note_to_chroma_by_max(notes)]
analyze_frame = []
# waveform
waveform = numpy.int8(waveform / (2**8))
# tempo analysis with tempo_frame
# if duration > 60:
# tempo_frame = numpy.concatenate(tempo_frame, axis=0)
# tempo = get_tempo(tempo_frame, tempo_fs)
# else: tempo = 0.0
tempo_frame = numpy.concatenate(tempo_frame, axis=0)
tempo = get_tempo(tempo_frame, tempo_fs)
xtempo = tempo
if xtempo <= 95:
xtempo = xtempo * 2.0
if xtempo == 0:
until_duration = duration
else:
until_duration = 60.0 / xtempo * best_highlight_duration_beat
# highlight analysis with highlight_raw_points
jump = 1
duratsum = []
highlight_length = until_duration * len(highlight_raw_points) / duration
duratsum = numpy.linspace(0, 0, len(highlight_raw_points))
# limit_factor = 1-(duration-60.0)/duration
# if limit_factor > 1.00: limit_factor = 0.00
# if limit_factor < 0.25: limit_factor = 0.25
# limithead = limit_factor
# limittail = limit_factor
# highlight_raw_points[:int(len(highlight_raw_points)*limithead)] = 0
# highlight_raw_points[-int(len(highlight_raw_points)*limittail):] = 0
limit_factor = 1 - (duration - 60.0) / duration
if limit_factor > 1.00:
limit_factor = 0.00
if limit_factor < 0.25:
limit_factor = 0.25
htlength = int(len(highlight_raw_points) * limit_factor)
window = numpy.hamming(htlength * 2)
highlight_raw_points[:
htlength] = highlight_raw_points[:
htlength] * window[:
htlength]
highlight_raw_points[
-htlength:] = highlight_raw_points[-htlength:] * window[htlength:]
for cnt in numpy.arange(0,
len(highlight_raw_points) - highlight_length,
jump):
thdata = numpy.mean(
highlight_raw_points[int(cnt):int(cnt + highlight_length)])
duratsum[int(cnt)] = thdata
pntadd = numpy.argmax(duratsum)
offset_time = 1.0 * jump * pntadd / (len(highlight_raw_points) / duration)
highlight = (offset_time, until_duration)
# autogain analysis in highlight period
autogain = 0.4
if find_gain:
if duration > 60:
autogain_analysis_length = highlight[1]
if duration - highlight[0] < autogain_analysis_length:
autogain_analysis_length = duration - highlight[0]
frame_length = fs * channel * autogain_analysis_length
byte_position = pybass.BASS_ChannelSeconds2Bytes(
hstream, highlight[0])
pybass.BASS_ChannelSetPosition(hstream, byte_position, False)
frame_raw = numpy.arange(frame_length, dtype=ctypes.c_short)
pybass.BASS_ChannelGetData(
hstream,
frame_raw.ctypes.data_as(ctypes.POINTER(ctypes.c_short)),
int(frame_length * 2))
mono_frame = frame_raw[::channel] / 32768.0
mono_frame = fir_filter(mono_frame,
lowcut=500,
highcut=fs / 2,
fs=fs,
order=15)
mono_frame = fir_filter(mono_frame,
lowcut=1000,
highcut=fs / 2,
fs=fs,
order=7)
mono_frame = fir_filter(mono_frame,
lowcut=1000,
highcut=fs / 2,
fs=fs,
order=5)
mono_frame = fir_filter(mono_frame,
lowcut=5000,
highcut=fs / 2,
fs=fs,
order=5)
if fs / 2 > 21000:
mono_frame = fir_filter(mono_frame,
lowcut=0,
highcut=20000,
fs=fs,
order=45)
mono_frame += fir_filter(
mono_frame, lowcut=15000, highcut=fs / 2, fs=fs,
order=5) * 0.5
rms = numpy.mean(mono_frame**2)**0.5 * 3
# spectrum = numpy.fft.fft(mono_frame, fs)
# spectrum = numpy.abs(spectrum[1:int(len(spectrum)/2)])
# pylab.plot(spectrum); pylab.show()
autogain = 0.14 / rms
# key analysis in highlight period
key = ''
if find_key:
chromagram, resolution = ([], 1.0)
note_freq_div = get_note_freq_div(resolution)
# note_window = get_note_window(fs, resolution, note_freq_div)
if xtempo == 0:
frame_length = int(fs * channel * 0.5)
else:
frame_length = int(fs * channel * (60.0 / xtempo))
offset_position, until_position = (highlight[0],
fs * channel * until_duration * 2)
if frame_length > total_frame_length:
frame_length = total_frame_length - 1
if offset_position + until_position > total_frame_length:
until_position = total_frame_length - offset_position
frame_raw = numpy.arange(frame_length, dtype=ctypes.c_short)
byte_position = pybass.BASS_ChannelSeconds2Bytes(
hstream, offset_position)
pybass.BASS_ChannelSetPosition(hstream, byte_position, False)
for cnt, frame_position in enumerate(
numpy.arange(0, total_frame_length - frame_length,
frame_length)):
pybass.BASS_ChannelGetData(
hstream,
frame_raw.ctypes.data_as(ctypes.POINTER(ctypes.c_short)),
int(frame_length * 2))
mono_frame = frame_raw[::channel] / 32768.0
spectrum = numpy.fft.fft(mono_frame, int(fs * resolution))
spectrum = numpy.abs(spectrum[1:int(len(spectrum) / 2)])
if find_key:
notes = spectrum_to_note_by_max(spectrum, note_freq_div)
chromagram += [note_to_chroma_by_max(notes)]
if (cnt + 1) * frame_length >= until_position:
break
scored_keys, key_scores, key_counts = ([], [0] * 24, [0] * 24)
for chroma in chromagram:
lag, score = get_chord_binaries_correlation_lag_score(chroma)
scored_keys += [lag]
key_counts[lag] += 1
key_scores[lag] += score
key_scores = numpy.array(key_scores)
max_key_scores = max(key_scores)
if max_key_scores == 0.0:
key = ''
else:
key_scores = key_scores / max_key_scores * 100
scored_key_idx = []
for i in range(1):
value, pnt = find_max(key_scores)
if value < 50:
break
scored_key_idx += [pnt[0]]
key_scores[pnt[0]] = 0
string_keys = []
for i in range(len(scored_key_idx) - 1, -1, -1):
if scored_key_idx[i] - 12 in scored_key_idx:
scored_key_idx.pop(i)
continue
elif scored_key_idx[i] + 12 in scored_key_idx:
scored_key_idx.pop(i)
continue
string_keys += [chord_idx_to_char(scored_key_idx[i])]
string_keys = ' or '.join(string_keys)
key = '%s' % (string_keys)
# chromagram = numpy.array(chromagram).T
# chromagram = numpy.flipud(chromagram)
# pylab.imshow(chromagram, interpolation='nearest')
# pylab.show() # pylab.grid(True)
# md5 = file2md5(upath)
tempo_type, key_type, save_tag = open_shelves(
('tempo_restict_type', 'key_format_type', 'auto_save_tag'),
PREFERENCE_DB)
if tempo_type is None or tempo_type == 0:
pass
elif tempo_type == 1 and tempo <= 95:
tempo = tempo * 2.0
elif tempo_type == 2 and tempo >= 120:
tempo = tempo * 0.5
if key_type is None:
key_type = 1
key = convert_chord_type(key, key_type)
if save_tag is None:
save_tag = False
if save_tag:
mutagen_mp3 = mutagen.mp3.MP3(path)
mutagen_mp3['TKEY'] = mutagen.id3.TKEY(encoding=3, text=[key])
mutagen_mp3['TBPM'] = mutagen.id3.TBPM(encoding=3, text=[tempo])
mutagen_mp3.save()
mfeats_data = MFEATS(mdx=mdx,
path=path,
key=key,
tempo=tempo,
duration=duration,
highlight=highlight,
waveform=waveform,
date=time.time(),
version=version,
channel=channel,
bit=bit,
error=0,
autogain=autogain)
# pybass.BASS_Free()
if pybass.BASS_ChannelIsActive(hstream) == 1:
pybass.BASS_StreamFree(hstream)
# print 'mfeats_single_finished: elapsed_time: %03.06fsec: %03.06fmsec/onesec'\
# % (time.time()-tic, (time.time()-tic)/duration*1000)
# return mfeats_data
if queue is not None:
queue.put(mfeats_data)
return
else:
return mfeats_data
