trial.py

#高速化のためのトライアル

import numpy as np
from PIL import Image
import moviepy.editor as mp
from pydub import AudioSegment
import cv2
from tqdm import tqdm 
import time

video_file = 'ito.mp4'
video = cv2.VideoCapture(video_file)

clip_input = mp.VideoFileClip(video_file)
#clip_input.audio.write_audiofile('audio.mp3')
sound_in = AudioSegment.from_file('audio.mp3', format="mp3")

frame_count = int(video.get(7))
frame_rate = int(video.get(5))
audio_rate = 1000 / frame_rate

width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT))

out_videofilename ='output.mp4'

# Set output video infomation.
fourcc = cv2.VideoWriter_fourcc(*'XVID')
#fourcc = cv2.VideoWriter_fourcc(*'MP4S')
vw = cv2.VideoWriter('result.mp4', fourcc, frame_rate, (width, height)) 

images_dir='img'

print(frame_count,frame_rate,audio_rate)
frame_count= frame_rate*50

#sounddata = sound.get_array_of_samples()
#print(sounddata)
sound_out = AudioSegment.empty()
volume=0
cnt=0
t1 = time.time() 
buf = np.zeros(frame_count)
for i in tqdm(range(0, frame_count)):
    #is_read, frame = video.read()
    #cv2.imwrite(images_dir+ "/" + str(i // frame_rate).zfill(6) + ".png", frame)
    flagment = sound_in[i*audio_rate:(i+1)*audio_rate]
    buf[i]=flagment.max
    if flagment.max > 500:
        cnt=2
    if cnt>0:
        cnt-=1
        sound_out += flagment
        #vw.write(frame) 
t2 = time.time() 
elapsed_time = t2-t1
print("elapsed time %04.04f"%elapsed_time)

#multi thread counting
import threading
t1 = time.time() 
buf = np.zeros(frame_count)
flm = dict()
print(buf)
workers = 10
end = 0
def counter(idx,len):
    global buf,end
    for idx in tqdm(range(i*len,i*len+len)):
        f = sound_in[idx*audio_rate:(idx+1)*audio_rate]
        buf[idx] = f.max
        flm[idx] = f
    end += 1

for i in tqdm(range(0, workers)):
    threading.Thread(target=counter, args=(i, frame_count//workers)).start()
t2 = time.time() 
elapsed_time = t2-t1
print("-----")
print(buf)
while end<workers:
    print(end)
    time.sleep(1)
print("elapsed time %04.04f"%elapsed_time)

for i in tqdm(range(0, frame_count)):
    is_read, frame = video.read()
    #cv2.imwrite(images_dir+ "/" + str(i // frame_rate).zfill(6) + ".png", frame)
    flagment = sound_in[i*audio_rate:(i+1)*audio_rate]
    if buf[i] > 500:
        cnt=2
    if cnt>0:
        cnt-=1
        sound_out += flm[i]
        vw.write(frame) 


sound_out.export("result.wav", format="wav")

import subprocess
subprocess.call("ffmpeg -i result.mp4 -i result.wav -c:v copy -c:a aac  output.mp4".split())