Skip to content

Instantly share code, notes, and snippets.

@kpj
Last active Jan 2, 2016
Embed
What would you like to do?
Non-working test
# requires ffmpeg, mpgtx
import sys, os, os.path, datetime
import numpy
from scipy.io.wavfile import read
# adjust this!
save_for_plot = False
volume_threshold = 250
if len(sys.argv) != 2:
sys.exit(1)
video_file = sys.argv[1] + ".mpg"
final_file = "final_" + video_file
# helper functions
def frmt(secs):
"""Turns seconds into suitable format
"""
return str(datetime.timedelta(seconds=round(secs)))
def aggregate_channel(chan):
"""Gathers information of one channel into one number
"""
return abs(numpy.average(chan))
def save_data(sample_rate, data):
with open('data', 'w') as fd:
for i, d in enumerate(data):
if i % sample_rate != 0:
continue
ac = aggregate_channel(d)
t = 0 if i == 0 else 1 / (sample_rate / i)
fd.write('%f %i\n' % (t, ac))
# convert video to mpg
if not os.path.exists(video_file):
os.system('ffmpeg -i "%s" "%s"' % (sys.argv[1], video_file))
# extract audio
audio_file = video_file + ".wav"
if not os.path.exists(audio_file):
os.system('ffmpeg -i "%s" -ab 160k -ac 2 -ar 44100 -vn "%s"' % (video_file, audio_file))
# read audio
sample_rate, data = read(audio_file)
# sample rate: # of samples per second
sample_num, channel_num = data.shape
# only save if wanted
if save_for_plot:
save_data(sample_rate, data)
sys.exit()
# normalize audio
#dtype = numpy.float64
#dtype = numpy.dtype(dtype)
#data = data.astype(dtype) / dtype.type(-numpy.iinfo(data.dtype).min)
# analyze audio
above = True
intersections = []
last_entries = []
le_len = 1000
# get intersections
for i, d in enumerate(data):
# calc time
t = 0 if i == 0 else 1 / (sample_rate / i)
# handle average list
last_entries.append(aggregate_channel(d))
if len(last_entries) > le_len:
last_entries.pop(0)
avg = numpy.average(last_entries)
if avg < volume_threshold:
if above:
# were above threshold in previous sample
intersections.append(t)
above = False
else:
if not above:
# were below threshold in last sample
intersections.append(t)
above = True
intersections.remove(0)
# get matching pairs
seqs = ""
i = 0
while i < len(intersections) - 1:
t1 = frmt(intersections[i])
t2 = frmt(intersections[i+1])
if t1 != t2:
seqs += "[%s-%s] " % (t1, t2)
i += 2
# cut video
print(seqs)
if os.path.exists(final_file):
os.remove(final_file)
os.system('mpgtx -j "%s" %s -o "%s"' % (video_file, seqs, final_file))
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment