Skip to content

Instantly share code, notes, and snippets.

@phelrine
Created November 18, 2016 07:27
Show Gist options
  • Save phelrine/0e7fcae91ecad851cbabf920579d1af6 to your computer and use it in GitHub Desktop.
Save phelrine/0e7fcae91ecad851cbabf920579d1af6 to your computer and use it in GitHub Desktop.
import sys
from pydub import AudioSegment, silence
import numpy as np
import matplotlib.pyplot as plt
SAMPLING_RATE = 44100
# THRESHOLD = 50
# CONDITION_INTERVAL = 100
THRESHOLD = 50
CONDITION_INTERVAL = 30
OFFSET = 1
def main():
audio_segment = AudioSegment.from_mp3(sys.argv[1])
start_segment = audio_segment[OFFSET * 1000:15 * 1000]
end_segment = audio_segment[-15 * 1000:-1000 * OFFSET]
start_offset = detect_silence(start_segment.get_array_of_samples(), threshold = 50)
end_offset = detect_silence(end_segment.get_array_of_samples(), threshold = 10, reversed = True)
print "%lf\t%lf" % (start_offset, end_offset)
def detect_silence(segment, threshold, reversed = False):
samples = np.array(segment)
channel_data = samples.reshape(len(samples) / 2, 2)
index = np.argmin(np.abs(np.average(channel_data, axis = 0)))
sample_width = SAMPLING_RATE * 0.01
sample_points = int(len(channel_data) / sample_width)
means = np.zeros(sample_points)
stds = np.zeros(sample_points)
for i in range(sample_points):
sample_index = i * sample_width
data = []
if reversed:
data = channel_data[int(sample_index - SAMPLING_RATE * 0.05):int(sample_index), index]
else:
data = channel_data[int(sample_index):int(sample_index + SAMPLING_RATE * 0.05), index]
if len(data) > 0:
means[i] = np.average(np.abs(data))
stds[i] = np.std(data)
sample_count = 0
for i in range(sample_points):
data = []
if reversed:
data = (stds[-(i + CONDITION_INTERVAL):-i] - np.min(stds)) > THRESHOLD
else:
data = (stds[i:i + CONDITION_INTERVAL] - np.min(stds)) > THRESHOLD
if len(data) > 0 and np.all(data):
sample_count = i
break
else:
sample_count = sample_points
# plt.subplot(3, 1, 1)
# plt.plot(means, color = 'g')
# plt.subplot(3, 1, 2)
# plt.plot((stds - np.min(stds))[0:200 * sample_width])
# plt.subplot(3, 1, 3)
# plt.plot(channel_data[:, index])
# plt.show()
return float(sample_count * sample_width) / SAMPLING_RATE + OFFSET
if __name__ == '__main__':
main()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment