def from_data(data, header=None):
"""build from data"""
if not isinstance(data, str):
raise TypeError('needs a sting as input!')
at = 0
# begin to build package
grp_idx, kind, nc, tf = unpack('<HcHH', data[at:at + 7])
data_lst = [grp_idx, kind, nc, tf]
tf_nc = tf * nc
at += 7
xcoors = sp.frombuffer(data[at:at + (nc * nc * (tf * 2 - 1) * 4)],
dtype=sp.float32)
xcoors.shape = (nc * nc, 2 * tf - 1)
data_lst.append(xcoors)
at += nc * nc * (tf * 2 - 1) * 4
cov = sp.frombuffer(data[at:at + (tf_nc * tf_nc * 4)],
dtype=sp.float32)
cov.shape = (tf_nc, tf_nc)
data_lst.append(cov)
return BS3CoveDataBlock(data_lst, header=header)
def svm_read_problem(self, data_file_name, return_scipy=False):
"""
svm_read_problem(data_file_name, return_scipy=False) -> [y, x], y: list, x: list of dictionary
svm_read_problem(data_file_name, return_scipy=True) -> [y, x], y: ndarray, x: csr_matrix
Read LIBSVM-format data from data_file_name and return labels y
and data instances x.
"""
if scipy != None and return_scipy:
prob_y = array('d')
prob_x = array('d')
row_ptr = array('l', [0])
col_idx = array('l')
else:
prob_y = []
prob_x = []
row_ptr = [0]
col_idx = []
indx_start = 1
for i, line in enumerate(open(data_file_name)):
line = line.split(None, 1)
# In case an instance with all zero features
if len(line) == 1: line += ['']
label, features = line
prob_y.append(float(label))
if scipy != None and return_scipy:
nz = 0
for e in features.split():
ind, val = e.split(":")
if ind == '0':
indx_start = 0
val = float(val)
if val != 0:
col_idx.append(int(ind) - indx_start)
prob_x.append(val)
nz += 1
row_ptr.append(row_ptr[-1] + nz)
else:
xi = {}
for e in features.split():
ind, val = e.split(":")
xi[int(ind)] = float(val)
prob_x += [xi]
if scipy != None and return_scipy:
prob_y = pd.DataFrame(scipy.frombuffer(prob_y, dtype='d'))
prob_x = scipy.frombuffer(prob_x, dtype='d')
col_idx = scipy.frombuffer(col_idx, dtype='l')
row_ptr = scipy.frombuffer(row_ptr, dtype='l')
prob_x = pd.DataFrame(sparse.csr_matrix((prob_x, col_idx, row_ptr)).todense())
return (prob_y, prob_x)
def svm_read_problem(data_file_name, return_scipy=False):
"""
svm_read_problem(data_file_name, return_scipy=False) -> [y, x], y: list, x: list of dictionary
svm_read_problem(data_file_name, return_scipy=True) -> [y, x], y: ndarray, x: csr_matrix
Read LIBSVM-format data from data_file_name and return labels y
and data instances x.
"""
if scipy != None and return_scipy:
prob_y = array('d')
prob_x = array('d')
row_ptr = array('l', [0])
col_idx = array('l')
else:
prob_y = []
prob_x = []
row_ptr = [0]
col_idx = []
indx_start = 1
for i, line in enumerate(open(data_file_name)):
line = line.split(None, 1)
# In case an instance with all zero features
if len(line) == 1: line += ['']
label, features = line
prob_y.append(float(label))
if scipy != None and return_scipy:
nz = 0
for e in features.split():
ind, val = e.split(":")
if ind == '0':
indx_start = 0
val = float(val)
if val != 0:
col_idx.append(int(ind)-indx_start)
prob_x.append(val)
nz += 1
row_ptr.append(row_ptr[-1]+nz)
else:
xi = {}
for e in features.split():
ind, val = e.split(":")
xi[int(ind)] = float(val)
prob_x += [xi]
if scipy != None and return_scipy:
prob_y = scipy.frombuffer(prob_y, dtype='d')
prob_x = scipy.frombuffer(prob_x, dtype='d')
col_idx = scipy.frombuffer(col_idx, dtype='l')
row_ptr = scipy.frombuffer(row_ptr, dtype='l')
prob_x = sparse.csr_matrix((prob_x, col_idx, row_ptr))
return (prob_y, prob_x)
def play(self):
self.set_wave()
self.set_stream()
self.data = self.wf.readframes(self.chunk)
self.changePlaySpeed(self.current_speed)
while len(self.data) > 0:
self.stream.write(self.data)
self.data = self.wf.readframes(self.chunk)
self.data = frombuffer(self.data, dtype="int16")
if self.task is None:
pass
elif self.task[0] == "pause":
self.task = None
self.pause()
elif self.task[0] == "stop":
self.task = None
self.stop()
elif self.task[0] == "change_speed":
print(f"curren_speed changed to: {self.task[1]}")
self.current_speed = self.task[1]
self.task = None
self.changePlaySpeed(self.current_speed)
print("play done")
self.isstart = False
def read_libsvm_format(
file_path: str) -> 'tuple[list[list[int]], sparse.csr_matrix]':
"""Read multi-label LIBSVM-format data.
Args:
file_path (str): Path to file.
Returns:
tuple[list[list[int]], sparse.csr_matrix]: A tuple of labels and features.
"""
def as_ints(str):
return [int(s) for s in str.split(',')]
prob_y = []
prob_x = array('d')
row_ptr = array('l', [0])
col_idx = array('l')
for i, line in enumerate(open(file_path)):
line = line.split(None, 1)
# In case an instance with all zero features
if len(line) == 1:
line += ['']
label, features = line
prob_y.append(as_ints(label))
nz = 0
for e in features.split():
ind, val = e.split(':')
val = float(val)
if val != 0:
col_idx.append(int(ind) - 1)
prob_x.append(val)
nz += 1
row_ptr.append(row_ptr[-1] + nz)
prob_x = scipy.frombuffer(prob_x, dtype='d')
col_idx = scipy.frombuffer(col_idx, dtype='l')
row_ptr = scipy.frombuffer(row_ptr, dtype='l')
prob_x = sparse.csr_matrix((prob_x, col_idx, row_ptr))
return (prob_y, prob_x)
def wavread(filename, normalize=False, dtype=float64):
win = wave.open(filename, 'r')
length = win.getnframes() * win.getsampwidth() * win.getnchannels()
nframes = win.getnframes()
x = frombuffer(win.readframes(length), dtype='c').view(typeinfo[win.getsampwidth()]).reshape(nframes, win.getnchannels()).astype(dtype)
win.close()
if normalize:
x /= abs(x).max()
else:
x = x / (2 ** (win.getsampwidth() * 8 - 1))
# print "==============================================12341234==="
return x, win.getframerate(), win.getsampwidth() * 8
def from_data(data, header=None):
"""build from data"""
if not isinstance(data, str):
raise TypeError('needs a sting as input!')
at = 0
# groups
group_lst = []
ngroup, = unpack('<H', data[at:at + 2])
at += 2
if ngroup > 0:
for _ in xrange(ngroup):
grp_idx, nc, tf, cl = unpack('<HHHH', data[at:at + 8])
at += 8
tf_nc = tf * nc
cov = sp.frombuffer(data[at:at + tf_nc * tf_nc * 4],
dtype=sp.float32)
at += tf_nc * tf_nc * 4
cov.shape = (tf_nc, tf_nc)
nunit, = unpack('<I', data[at:at + 4])
at += 4
unit_lst = []
if nunit > 0:
for _ in xrange(nunit):
filt = sp.frombuffer(
data[at:at + tf_nc * 4],
dtype=sp.float32
).reshape(tf, nc).T
at += tf_nc * 4
temp = sp.frombuffer(
data[at:at + tf_nc * 4],
dtype=sp.float32
).reshape(tf, nc).T
at += tf_nc * 4
snr, active, u1, u2 = unpack('<fBHH', data[at:at + 9])
at += 9
unit_lst.append((filt, temp, snr, active, u1, u2))
group_lst.append((grp_idx, nc, tf, cl, cov, unit_lst))
return BS3SortSetupBlock(group_lst, header=header)
def wavread_old(fname, dtype='float'):
"""
Read wave file
Parameters:
fname: string
input filename
Returns:
outData: ndarray
wave data of fname
fs: float
sampling rate of fname
bit: int
bit depth of fname
"""
wp = wave.open(fname, 'rb')
fs = wp.getframerate()
bit = wp.getsampwidth() * 8
ch = wp.getnchannels()
length = wp.getnframes()
if dtype == 'float':
scaling = 32768.0
elif dtype == 'int':
scaling = 1
else:
scaling = 32768.0
str_data = wp.readframes(wp.getnframes())
if bit == 16:
data = sp.frombuffer(str_data, 'int16') / scaling # 24bit wavはどうすればいい？
elif bit == 24:
#data = sp.fromstring(strData, sp.int32) # 24bit wavはどうすればいい？
data = sp.zeros(wp.getnframes())
if (ch == 2):
#lData = sp.int32(data[::2])
#rData = sp.int32(data[1::2])
l_data = data[::2]
r_data = data[1::2]
out_data = sp.array([l_data, r_data])
else:
#lData = sp.int32(data)
l_data = data
out_data = sp.array(l_data)
return out_data, fs, bit
def fill_image(self, i, l, use_binarization=True, equalize=False):
text = l[-1]
if len(text) > self.word_size:
return None
gray = int(l[2])
box = [int(l[3]), int(l[4]), int(l[5]), int(l[6])]
word_id = l[0].split("-")
line_id="-".join(word_id[:3])
line_box = self.lines[line_id]
word_dy = box[1]-line_box[1]
if word_dy < 0:
return None
scale = self.scale*random.uniform(0.9, 1.1)
new_height = floor(box[3]*scale)
if (new_height>self.line_height):
return None
new_width = floor(box[2]*scale)
if new_width < 2 or new_height < 2:
return None
f_name = self.word_file(l)
data = self.word_images[l[0]]
data = sp.frombuffer(data, sp.uint8)
try:
img = cv2.imdecode(data, 0)
if equalize:
img = cv2.equalizeHist(img,gray)
elif use_binarization:
_, img = cv2.threshold(img,gray,255,cv2.THRESH_BINARY)
img=cv2.resize(img, (new_width, new_height), cv2.INTER_LANCZOS4)
word_dy = 128 - floor(word_dy*self.scale) - new_height + floor(random.uniform(-self.rand_y, self.rand_y))
dx = self.dx + floor(random.uniform(0, self.rand_x))
max_x = dx+new_width
if max_x > self.line_width:
new_width -= max_x - self.line_width
self.tmp[i, word_dy:word_dy+new_height, 0+dx:dx+new_width]=img
except:
return None
w = l[-1]
W = self.encode_word(w)
return W
def separate_wav(self, cut_interval, shift_time):
# frames = int(self.s_info.channels * self.s_info.f_rate)
frames = self.s_info.f_rate
shift_frames = int(frames * shift_time)
chunk_size = int(frames * cut_interval)
sample = frombuffer(self.s_info.data, dtype=int16)
sample_len = len(sample)
def calc_begin_end(offset: int) -> Tuple[int, int]:
begin = offset * shift_frames
end = begin + chunk_size
return begin, end
delimiter = ((sample_len - frames) // shift_frames) + 1
for i in range(delimiter):
begin, end = calc_begin_end(offset=i)
yield sample[begin:end]
def _get_echan(self, idx):
"""yields an event channel as ndarray
:type idx: int
:param idx: channel id.
"""
# checks
if idx not in self.echan_header:
raise IndexError('no data for this channel: %s' % idx)
# get data
self.fp.seek(self.echan_header[idx].data_offset, 0)
byte_data = self.fp.read(self.echan_header[idx].n_sample * 4)
# return
if len(byte_data) == 0:
return sp.array([], dtype=sp.int32)
return sp.frombuffer(byte_data, dtype=sp.int32)
def _get_echan(self, idx):
"""yields an event channel as ndarray
:type idx: int
:param idx: channel id.
"""
# checks
if idx not in self.echan_header:
raise IndexError('no data for this channel: %s' % idx)
# get data
self.fp.seek(self.echan_header[idx].data_offset, 0)
byte_data = self.fp.read(self.echan_header[idx].n_sample * 4)
# return
if len(byte_data) == 0:
return sp.array([], dtype=sp.int32)
return sp.frombuffer(byte_data, dtype=sp.int32)
def _get_achan(self, idx):
"""yields an analog channel as ndarray
:type idx: int
:param idx: channel id
"""
# checks
if idx >= self.max_achan:
raise IndexError('not a valid channel: %s' % idx)
if idx not in self.achan_header:
raise IndexError('no data for this channel: %s' % idx)
# get data
self.fp.seek(self.achan_header[idx].data_offset)
byte_data = self.fp.read(self.achan_header[idx].n_sample * 2)
# return
if len(byte_data) == 0:
return sp.array([], dtype=sp.int16)
return sp.frombuffer(byte_data, dtype=sp.int16)
def from_data(data, header=None):
"""build from data"""
if not isinstance(data, str):
raise TypeError('needs a sting as input!')
at = 0
# events
event_lst = []
nevent, = unpack('<I', data[at:at + 4])
at += 4
if nevent > 0:
for _ in xrange(nevent):
gid, uid, tv, nc, ns = unpack('<HIQHH', data[at:at + 18])
at += 18
wf = sp.frombuffer(
data[at:at + ns * nc * 2],
dtype=sp.int16
).reshape(nc, ns).T
at += ns * nc * 2
event_lst.append((gid, uid, tv, nc, ns, wf))
return BS3WaveDataBlock(event_lst, header=header)
def read_mhd(mhd_path):
""" reads the data from an mhd file and returns a np array. The data
type to read is specified from the tifffile module by the tiff reading
capabilities. Based on code from the pirt library.
see: https://bitbucket.org/almarklein/pirt"""
# Load description from mhd file
mhd = open(mhd_path, 'r').read()
# Get data filename and load raw data
raw_path = re.findall('ElementDataFile = (.+)', mhd)[0]
# if the path in the mhd is not an absolute path, make it one
if raw_path[0] != '/':
raw_path = os.path.join(os.path.dirname(mhd_path),
os.path.basename(raw_path))
# get dimensions
dimensions = sp.int16(re.findall('DimSize = (.+)', mhd)[0].split())
# get correct datatype mapping
mhd_dtype = re.findall('ElementType = (.+)', mhd)[0]
dtype = get_np_dtype(mhd_dtype)
# read and reshape
fh = open(raw_path, 'rb')
data = sp.frombuffer(fh.read(), dtype=dtype)
if len(dimensions) == 2:
data_reshape = sp.reshape(data, (dimensions[1], dimensions[0]))
data_reshape = data_reshape.T
if len(dimensions) == 3: # FIXME
sys.exit()
data_reshape = sp.reshape(
data, (dimensions[2], dimensions[1], dimensions[0]))
return data_reshape
def split_mp3_to_wav_segment(filename, sampling_rate, mp3_length, num_segment,
index):
wavf = str(filename) + '.wav'
wr = wave.open(wavf, 'r')
ch = wr.getnchannels()
width = wr.getsampwidth()
fr = wr.getframerate()
fn = wr.getnframes()
total_time = 1.0 * fn / fr
t = mp3_length / num_segment
frames = int(ch * fr * t)
data = wr.readframes(wr.getnframes())
wr.close()
X = frombuffer(data, dtype=int16)
if total_time < mp3_length:
print('mp3_length must be shorter than total_time')
for j in range(num_segment):
split = df.split[index]
clip_id = df.clip_id[index]
outf = 'wav_segment/' + split + '_' + str(clip_id) + '_' + str(
j + 1) + '.wav'
x = math.floor(sampling_rate * (total_time - mp3_length) / 2)
start_cut = j * frames + x
end_cut = j * frames + frames + x
Y = X[start_cut:end_cut]
outd = struct.pack("h" * len(Y), *Y)
ww = wave.open(outf, 'w')
ww.setnchannels(ch)
ww.setsampwidth(width)
ww.setframerate(fr)
ww.writeframes(outd)
ww.close()
def read_mhd(mhd_path):
""" reads the data from an mhd file and returns a np array. The data
type to read is specified from the tifffile module by the tiff reading
capabilities. Based on code from the pirt library.
see: https://bitbucket.org/almarklein/pirt"""
# Load description from mhd file
mhd = open(mhd_path,'r').read()
# Get data filename and load raw data
raw_path = re.findall('ElementDataFile = (.+)',mhd)[0]
# if the path in the mhd is not an absolute path, make it one
if raw_path[0] != '/':
raw_path = os.path.join(os.path.dirname(mhd_path),os.path.basename(raw_path))
# get dimensions
dimensions = sp.int16(re.findall('DimSize = (.+)',mhd)[0].split())
# get correct datatype mapping
mhd_dtype = re.findall('ElementType = (.+)',mhd)[0]
dtype = get_np_dtype(mhd_dtype)
# read and reshape
fh = open(raw_path, 'rb')
data = sp.frombuffer(fh.read(),dtype = dtype)
if len(dimensions) == 2:
data_reshape = sp.reshape(data,(dimensions[1],dimensions[0]))
data_reshape = data_reshape.T
if len(dimensions) == 3: # FIXME
sys.exit()
data_reshape = sp.reshape(data,(dimensions[2],dimensions[1],dimensions[0]))
return data_reshape
('S1_phi', None),
('S1_eta', None),
('S1_theta', None),
)
print "... going to fetch data from tree"
##### copy the data to memory. This is super fast, excellent!!
data_buf = [None] * len(data_format)
tIn.SetEstimate(nevents + 1)
for idx, df in enumerate(data_format):
print '... reading data in', df[0]
tIn.Draw(df[0], "", "goff", nevents, 0)
temp = tIn.GetV1()
data_buf[idx] = copy.deepcopy(
scipy.frombuffer(buffer=temp, dtype='double', count=nevents))
data = np.asarray(data_buf)
dataframe = pd.DataFrame(data, index=[x[0] for x in data_format])
pd.set_option('expand_frame_repr', False)
pd.options.display.max_rows = 10
print "... data moved to memory"
# print data
# print dataframe
### transpose, so that row ==> evt index, column ==> property
dataframe = dataframe.transpose()
print "... dataset transposed"
print dataframe
else:
## first determine average for each
##### copy the data to memory. This is super fast, excellent!!
nevents = tIn.GetEntries()
tIn.SetEstimate(nevents + 1)
avgs = [0, 0, 0, 0]
for idx in range(4):
print '... making averages', idx + 1
thecut = cut.format("S%i" % (idx + 1)) + " && has_S%i" % (idx + 1)
print thecut
tIn.Draw(expr.format("S%i") % (idx + 1), thecut, "goff", nevents, 0)
temp = tIn.GetV1()
nselected = tIn.GetEntries(thecut)
print nselected
data_buf = copy.deepcopy(
scipy.frombuffer(buffer=temp, dtype='double', count=nselected))
data = np.asarray(data_buf)
# print data
# print min(data), max(data)
avgs[idx] = np.average(data)
print " >> avg ", avgs[idx]
h_S1 = make_histogram(tIn,
expr.format("S1"),
cut.format("S1") + " && has_S1",
hbounds,
'hscat_S1',
xshift=-1. * avgs[0]) ## to be centred at 0
h_S2 = make_histogram(tIn,
expr.format("S2"),
cut.format("S2") + " && has_S2",
# check if the word satisfies all thresholds
if wordConfi > confidence_threshold and (
wordT1 - wordT0) / wordCount > length_threshold:
confidences[wordPy] = wordConfi
timestamps[wordPy] = [wordT0, wordT1]
# extract basic information of the wave file
audio = wave.open(wav_file, 'r')
ch = audio.getnchannels()
width = audio.getsampwidth()
fr = audio.getframerate()
fn = audio.getnframes()
data = audio.readframes(fn)
audioContent = frombuffer(data, dtype=int16)
# split the wave file
if not os.path.exists('./output'):
os.makedirs('./output')
for name in timestamps:
segment = audioContent[int(timestamps[name][0] * fr *
ch):int(timestamps[name][1] * fr * ch)]
outd = struct.pack("h" * len(segment), *segment)
ww = wave.open('./output/' + name + '.wav', 'w')
ww.setnchannels(ch)
ww.setsampwidth(width)
ww.setframerate(fr)
ww.writeframes(outd)
ww.close()
# ('genjet_eta' , 'double'),
# ('genjet_phi' , 'double'),
# ('genjet_mass' , 'double'),
)
print "... going to fetch data from tree"
##### copy the data to memory. This is super fast, excellent!!
data_buf = [None] * len(data_format)
tIn.SetEstimate(nevents + 1)
for idx, df in enumerate(data_format):
print '... reading data in', df[0]
tIn.Draw(df[0], "", "goff", nevents, 0)
temp = tIn.GetV1()
data_buf[idx] = copy.deepcopy(
scipy.frombuffer(buffer=temp, dtype=df[1], count=nevents))
print "... done"
data = np.asarray(data_buf)
df = pd.DataFrame(data, index=[x[0] for x in data_format])
df = df.transpose()
## graphics options for display
pd.set_option('expand_frame_repr', False)
pd.options.display.max_rows = 10
# print df
# convert jet flav column to integer type
df['jet_flav'] = df['jet_flav'].astype(int)
# use abs eta of jet
def cut_wav(filename, time):
wav_file = f'{filename}.wav'
wr = wave.open(wav_file, 'r')
ch = wr.getnchannels()
width = wr.getsampwidth()
fr = wr.getframerate()
fn = wr.getnframes()
total_time = 1.0 * fn / fr
integer = math.floor(total_time)
# t = int(time)
t = time
# フレームに応じて図の横幅が変わるため、一定値にする
frames = int(ch * fr * t)
num_cut = int(integer // t)
print(f'channle: {ch}')
print(f'sample width: {width}')
print(f'frame rate: {fr}')
print(f'frame num: {fn}')
print(f'params: {wr.getparams()}')
print(f'total time: {total_time}')
print(f'total time(int): {integer}')
print(f'time: {t}')
print(f'frames: {frames}')
print(f'number of cut: {num_cut}')
data = wr.readframes(wr.getnframes())
wr.close()
X = frombuffer(data, dtype=int16)
out_file_path = f'{export_dir}/{filename}'
if not os.path.exists(out_file_path):
os.makedirs(out_file_path)
end_condition = frames * num_cut
filename = os.path.basename(filename)
frames = 10000 # 横幅を合わせるための
for i in range(num_cut):
print(i)
out_file = f'{out_file_path}/{filename}_{i}.wav'
start_cut = i * frames
end_cut = (i + 10) * frames
# end_cut = (i + 10) * frames + frames
# print(start_cut, end_cut)
# print()
if end_cut > end_condition:
print(' over frames')
return
Y = X[start_cut:end_cut]
out_date = struct.pack('h' * len(Y), *Y)
# output
with wave.open(out_file, 'w') as ww:
ww.setnchannels(ch)
ww.setsampwidth(width)
ww.setframerate(fr)
ww.writeframes(out_date)
OnePlotAxesX.spines['top'].set_visible(False) OnePlotAxesY.spines['right'].set_visible(False) OnePlotAxes.tick_params(axis='y', direction='out') OnePlotAxes.set_xlabel(var1label) OnePlotAxes.set_ylabel(var2label) N = mytree.GetEntries() corrhist = mytree.Draw(vartreeexp+">>htemp111%s(%d,%f,%f,%d,%f,%f)"% (MCSample,var1limits[2],var1limits[0],var1limits[1],var2limits[2],var2limits[0],var2limits[1]), selection="%f"%(MCScales[MCSample]), create_hist=True, options="goff") # mytree.GetSelectedRows() y = mytree.GetV1() y = copy.deepcopy(scipy.frombuffer(buffer=y,dtype='double',count=1000000 )) x = mytree.GetV2() x = copy.deepcopy(scipy.frombuffer(buffer=x,dtype='double',count=1000000 )) w = mytree.GetW() w = copy.deepcopy(scipy.frombuffer(buffer=w,dtype='double',count=1000000 )) # print w[:100] # print w[-100:] # plt.show() # code.interact(local=locals()) # Let's add on other samples that are related for (tmpMCSample,tmpNorm, tmpLabel, tmpFirstInCombinedSample) in MCSamples[jSample+1:]: if tmpFirstInCombinedSample:
def numpy_from_array(v):
# return scipy.frombuffer(v.get_obj())
return scipy.frombuffer(v.get_obj())
wavfile = "./futta-dream.wav"
# WAVファイルを開く
wr = wave.open(wavfile, "rb")
# WAVファイルの情報を表示（別にいらん）
print("Channel num : ", wr.getnchannels())
print("Sample size : ", wr.getsampwidth())
print("Sampling rate : ", wr.getframerate())
print("Frame num : ", wr.getnframes())
print("Prams : ", wr.getparams())
print("Sec : ", float(wr.getnframes()) / wr.getframerate())
# データの読み込み
data = wr.readframes(wr.getnframes())
# 文字型から数値型に
num_data = frombuffer(data, dtype=int16)
if (wr.getnchannels() == 2):
# 左チャンネル
left = num_data[::2]
# 右チャンネル
right = num_data[1::2]
wr.close()
plt.plot(left)
plt.show()
