def getSignalFromFile(filename):
filepath = os.path.join(WAV_PATH, filename)
framerate, signal = scipy_read(filepath)
assert (framerate == Parameters.sampling_rate)
assert (signal.shape[1] == Parameters.n_channels)
return signal
def __init__(self, path: str, hop_size: int):
super().__init__(hop_size)
self.samplerate, self.data = scipy_read(path)
self.duration_s = self.data.shape[0] / self.samplerate
df = pd.concat(df, axis=0, ignore_index=True).reset_index(drop=True)
p.close()
p.join()
print('After Shape: ', df.shape)
return df
#if not os.path.exists('../input/train/audio/silence'): #use on your own device to add files and use for noise in class files
# os.makedirs('../input/train/audio/silence')
samples = pd.DataFrame(
glob.glob('../input/train/audio/_background_noise_/**.wav'),
columns=['path'])
n = 10000
for w in samples.path:
rate, wav = scipy_read(w)
print(w, rate, wav.shape)
for i in range(0, len(wav) - 16000, 16000):
#scipy_write('../input/train/audio/silence/' + str(n) + '.wav', rate, wav[i:i+16000])
n += 1
audio_classes = [
'yes', 'no', 'up', 'down', 'left', 'right', 'on', 'off', 'stop', 'go',
'silence', 'unknown'
]
print('train...')
train = pd.DataFrame(glob.glob('../input/train/audio/**/**.wav'),
columns=['path'])
train['label'] = train['path'].map(lambda x: x.split('/')[4])
train['label'] = train['label'].map(lambda x: x
if x in audio_classes else 'unknown')
def fscipy_read(path):
try:
return scipy_read(path)[1]
except:
print(path)
return
# were used during training
# Number of MFCC features to use
N_FEATURES = 26
# Size of the context window used for producing timesteps in the input vector
N_CONTEXT = 9
ds = Model(lm_binary, N_FEATURES, N_CONTEXT, alphabet, BEAM_WIDTH)
ds.enableDecoderWithLM(alphabet, lm_binary, trie, LM_WEIGHT, WORD_COUNT_WEIGHT,
VALID_WORD_COUNT_WEIGHT)
# #python to batch process
# import subprocess
#
# for index, i in enumerate(prediction.fname):
# argument = output_graph +" "+ files[i]+" " + alphabet+" " + lm_binary+" " + trie
#
#
# with subprocess.Popen(['deepspeech', argument], shell=True,stdout=subprocess.PIPE) as item:
# for line in item.stdout:
# print (line)
for index, i in enumerate(prediction.fname):
with scipy_read(files[i]) as a:
fs, audio = a
#get the inference and replace it inside the pandas dataframe
prediction.set_value(index, 'label', ds.stt(audio, fs))
#save the prediction
prediction.to_csv("predictions.csv")