def move2(dir, n):
global loc2
track2[loc2].remove(char)
loc2 = loc2 + dir*n
loc2 = envelope(loc2, 0, lastind(track2))
track2[loc2].append(char)
def move2(dir, n):
global loc2
track2[loc2].remove(char)
loc2 = loc2 + dir * n
loc2 = envelope(loc2, 0, lastind(track2))
track2[loc2].append(char)
def move(dir, n):
"""Move `n` times in `dir` direction."""
global loc
track[loc] = blank
loc = envelope(loc + dir*n, 0, lastind(track))
track[loc] = char
def move(dir, n):
"""Move `n` times in `dir` direction."""
global loc
track[loc] = blank
loc = envelope(loc + dir * n, 0, lastind(track))
track[loc] = char
def process_signal(filename):
signal, sample_rate = librosa.load(p.join(DATA_DIR, filename),
sr=DOWNSAMPLED_SAMPLE_RATE)
rolling_mean_signal = envelope(signal, sample_rate, DATA_CHUNKS_PER_SECOND)
noise_mask = rolling_mean_signal > NOISE_AMPLITUDE_THRESHOLD
wavfile.write(filename=p.join(DEST_DIR, filename),
rate=DOWNSAMPLED_SAMPLE_RATE,
data=signal[noise_mask])
def convert_wav2npy(arg_tuple):
path = arg_tuple[0]
save_dir = arg_tuple[1]
sample_rate = arg_tuple[2]
denoise = arg_tuple[3]
x = librosa.load(path, sr=sample_rate, mono=True)[0]
if denoise:
try:
mask, env = utils.envelope(x, sample_rate, threshold=0.25)
x = nr.reduce_noise(audio_clip=x,
noise_clip=x[np.logical_not(mask)],
verbose=False)
except ValueError:
print("=" * 10)
print(path)
print(x)
# raise
ebird_code = path.parent.name
write_path = Path(save_dir) / f"{ebird_code}/{path.stem}.npy"
print(f"write to {write_path}")
# sf.write(write_path, x, sample_rate)
np.save(write_path, x)
def execute_flow(flow):
if flow not in get_flows():
return envelope('unknown flow {}'.format(flow))
settings = merge_settings(flow)
if request.method not in settings.get('http_methods'):
return envelope('invalid http method', success=False)
# read the args from request type data
params = request.form if request.method == 'POST' else request.args
if params.get('token') != settings.get('token'):
return envelope('incorrect token', success=False)
text = params.get('text')
if not text:
return envelope('empty text', success=False)
if not settings.get('activate_flow_re').search(text):
return envelope('no activation phrase found - ignored')
fields = dict(
(field, get_field_value(field, settings['selectors'][field], text))
for field in settings['selectors'].keys()
)
values = format_fields(fields, settings['fields'])
try:
update_spreadsheet(values, settings)
except Exception as e:
return envelope('error updating spreadsheet - '
'{} {}'.format(type(e).__name__, str(e)),
success=False)
return envelope('data successfully saved')
# Starting from the final position in the burn-in chain, sample for nsteps.
sampler.run_mcmc(pos, nsteps, rstate0=state)
###############################################################################
# statistical inference from sampling the loglikelihood parameter space
###############################################################################
# instruction to save the result of sampling:
# - the coordinates of the sampled points in the chain
# - the corresponding logprobability
np.save('results/{0}/{1}/samples_{0}_{1}'.format(ker, dwarf), sampler.chain)
np.save('results/{0}/{1}/lnprobs_{0}_{1}'.format(ker, dwarf),
sampler.lnprobability)
# instruction to flatten the chain to successively
# envelope the results along the desired direction
flatsamples = sampler.flatchain
flatlnprobs = sampler.flatlnprobability
# envelope lowermost -lnlikelihood values over sampled J range
# to obtain the profile of another variable, change 'param' argument
Penv, Senv, Lenv = envelope(flatsamples, flatlnprobs, param=2)
# print results of envelope on screen
print 'results'
for P, S, L in zip(Penv, Senv, Lenv):
print P, S, L
def get_data(save_locally = False,
sub_segments = True, num_segments=5,
cleaning = True, threshold = 0.005):
"""
Get the data
Adjust parameters :
# save_locally : save the dataframe
# sub_segments : split the audio into segments
# num_segments : number of segments
# cleaning : whether to apply a enveloppe mask
# threshold : threshold of the enveloppe
"""
arr_features=[]
# Get data path
path = os.path.join(os.getcwd(),'..','data')
# Get list of genres
genres = [f for f in os.listdir(path)]
genres = genres
for idx,genre in tqdm(enumerate(genres),total=len(genres)):
# Get genre pathing
genre_path = os.path.join(path,genre)
for fname in os.listdir(genre_path):
# Get file pathing
file_path = os.path.join(genre_path,fname)
y, sr = librosa.load(file_path, duration=30)
# Cleaning option
if cleaning == True :
mask = envelope(y,sr,threshold)
y = y[mask]
# Split songs into sub segments
if sub_segments == True :
# Get subsegments
track_duration = round(len(y)/sr)
samples_per_track = sr * track_duration
samples_per_segment = int(samples_per_track / num_segments)
for d in range(num_segments):
start = samples_per_segment * d
finish = start + samples_per_segment
new_y = y[start:finish]
# Get features
dict_features=extract_features(y=new_y,sr=sr)
# Get label
dict_features['label']=idx
# Keep name
dict_features['id']=f"{fname}_{d}"
# Total track duration
dict_features['total_duration']=track_duration
dict_features['sub_duration']= round((finish-start)/sr)
# Append to list
arr_features.append(dict_features)
# Don't split songs into subsegment
else:
# Get features
dict_features=extract_features(y=y,sr=sr)
# Get label
dict_features['label']=idx
# Keep name
dict_features['id']=f"{fname}"
# Append to list
arr_features.append(dict_features)
# Create a dataframe with the features
df=pd.DataFrame(data=arr_features)
# ----------------------------------------------------------
# Print final output details
# ----------------------------------------------------------
if cleaning == True :
print('# Envelope used for data cleaning')
if sub_segments == True:
print(f'# Split each sound into {num_segments} sub segments')
else :
print('# Sounds not split into sub segments')
if save_locally == False:
print('# Dataset generated but not saved')
print('# Shape of the dataset',df.shape)
else:
print('# Dataset generated')
print('# Shape of the dataset',df.shape)
df.to_csv(os.path.join(os.getcwd(),'..','dataset','clean_df.csv'),index=False)
print('# Dataset saved')
return df
