def create_tsm(name, channels, speed):
"""Create a TSM object given the method name and its parameters."""
if name == "ola":
return ola(channels, speed)
if name == "wsola":
return wsola(channels, speed)
if name == "phasevocoder":
return phasevocoder(channels, speed, phase_locking=PhaseLocking.NONE)
if name == "phasevocoder_identity":
return phasevocoder(channels, speed,
phase_locking=PhaseLocking.IDENTITY)
raise ValueError("unknown TSM method: {}".format(name))
def speed(self, speed):
"""
Adjusts speed to given percentage without changing pitch
:param speed: Percentage to increase/decrease speed without changing
pitch
:type speed: float
"""
if speed != 1:
logger.info("Setting speed to %f", speed)
logger.debug("Export file to BytesIO")
wav_in = BytesIO()
wav_in = self._segment.export(wav_in, format="wav")
wav_in.seek(0)
logger.debug("Initializing reader and writer")
with WavReader(wav_in) as reader:
wav_out = BytesIO()
with WavWriter(wav_out, reader.channels,
reader.samplerate) as writer:
logger.debug("Adjusting speed with vocoder")
tsm = phasevocoder(reader.channels, speed=speed)
tsm.run(reader, writer)
logger.debug("Reload audio segment")
wav_out.seek(44) # skip metadata and start at first sample
self._segment = AudioSegment.from_raw(
wav_out,
sample_width=self._segment.sample_width,
channels=self._segment.channels,
frame_rate=self._segment.frame_rate,
)
def change_pitch(self, source_filepath):
import pydub
from audiotsm.io.wav import WavReader, WavWriter
from audiotsm import phasevocoder
if abs(self.octaves) > 0.1:
_, sampled_filepath = tempfile.mkstemp(suffix='.wav', dir=self.temp_dir)
sound = pydub.AudioSegment.from_mp3(source_filepath)
sample_rate = int(sound.frame_rate * (2**self.octaves))
modified = sound._spawn(sound.raw_data, overrides={"frame_rate":sample_rate})
modified.export(sampled_filepath, format="wav")
else:
sampled_filepath = source_filepath
if abs(self.speed - 1) > 0.1:
#output_filepath = f"{os.path.basename(source_filepath)}{self.oct_str}_{self.speed}.wav"
_, output_filepath = tempfile.mkstemp(suffix='.wav', dir=self.temp_dir)
with WavReader(sampled_filepath) as reader:
with WavWriter(output_filepath, reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=self.speed*(sound.frame_rate/sample_rate))
tsm.run(reader, writer)
return output_filepath
else:
return sampled_filepath
def create_tsm(name, channels, parameters):
"""Create a TSM object given the method name and its parameters."""
if name == "ola":
return ola(channels, **parameters)
if name == "wsola":
return wsola(channels, **parameters)
if name == "phasevocoder":
return phasevocoder(channels, **parameters)
raise ValueError("unknown TSM method: {}".format(name))
def create_tsm(self, channels):
parameters = {}
if self.frame_length > 0:
parameters['frame_length'] = self.frame_length
if self.synthesis_hop > 0:
parameters['synthesis_hop'] = self.synthesis_hop
if self.phase_locking >= 0:
parameters['phase_locking'] = self.phase_locking
return phasevocoder(channels, **parameters)
def 音频变速(wav音频数据列表, 声道数, 采样率, 目标速度, 临时文件夹):
if 目标速度 == 1.0:
return wav音频数据列表
if 查找可执行程序('soundstretch') != None:
内存音频二进制缓存区 = io.BytesIO()
fd, soundstretch临时输出文件 = tempfile.mkstemp()
os.close(fd)
wavfile.write(内存音频二进制缓存区, 采样率, wav音频数据列表)
变速命令 = f'soundstretch stdin "{soundstretch临时输出文件}" -tempo={(目标速度 - 1) * 100}'
变速线程 = subprocess.Popen(变速命令, stdout=subprocess.PIPE, stdin=subprocess.PIPE, stderr=subprocess.PIPE)
变速线程.communicate(内存音频二进制缓存区.getvalue())
try:
采样率, 音频区间处理后的数据 = wavfile.read(soundstretch临时输出文件)
except Exception as e:
出错时间 = int(time.time())
fd, 原始数据存放位置 = tempfile.mkstemp(dir=临时文件夹, prefix=f'原始-{出错时间}-', suffix='.wav')
os.close(fd)
wavfile.write(原始数据存放位置, 采样率, wav音频数据列表)
fd, 出错文件 = tempfile.mkstemp(dir=临时文件夹, prefix=f'变速-{出错时间}-', suffix='.wav')
os.close(fd)
try:
copy(soundstretch临时输出文件, 出错文件)
except:
...
fd, soundstretch临时输出文件 = tempfile.mkstemp(dir=临时文件夹, prefix=f'变速-{出错时间}-', suffix='.wav')
os.close(fd)
变速命令 = f'soundstretch stdin "{soundstretch临时输出文件}" -tempo={(目标速度 - 1) * 100}'
变速线程 = subprocess.Popen(变速命令, stdout=subprocess.PIPE, stdin=subprocess.PIPE, stderr=subprocess.PIPE)
变速线程.communicate(内存音频二进制缓存区.getvalue())
print(f'Soundstretch 音频变速出错了，请前往查看详情\n 原始音频数据：{原始数据存放位置} \n 变速音频数据：{soundstretch临时输出文件}\n')
print(f'出错的音频信息：\n 音频采样数：{len(wav音频数据列表)}\n 目标速度：{目标速度}\n 目标采样数：{len(wav音频数据列表) / 目标速度}')
return wav音频数据列表
os.remove(soundstretch临时输出文件)
else:
print(
'检测到没有安装 SoundTouch 的 soundstretch，所以使用 phasevocoder 的音频变速方法。建议到 http://www.surina.net/soundtouch 下载系统对应的 soundstretch，放到系统环境变量下，可以获得更好的音频变速效果\n')
sFile = io.BytesIO()
wavfile.write(sFile, 采样率, wav音频数据列表)
sFile = io.BytesIO(sFile.getvalue())
eFile = io.BytesIO()
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=目标速度)
tsm.run(reader, writer)
_, 音频区间处理后的数据 = wavfile.read(io.BytesIO(eFile.getvalue()))
return 音频区间处理后的数据
def change_bar_speed(self, audio_slice_id, target_bpm=120.0):
if not os.path.isdir(c.LF_CH_BPM + self._audio_id):
try:
os.mkdir(c.LF_CH_BPM + self._audio_id)
except FileExistsError:
pass
else:
if os.path.isfile(c.LF_CH_BPM + self._audio_id + "/" +
audio_slice_id + ".wav"):
return 0
bar_bpm = 60.00 / (
(self.beat_track[int(audio_slice_id.split("_")[1]) + 1] -
self.beat_track[int(audio_slice_id.split("_")[1])]) / 8)
with WavReader("{}{}/{}.wav".format(c.LF_SLICE, self._audio_id,
audio_slice_id)) as r:
with WavWriter(
"{}{}/{}.wav".format(c.LF_CH_BPM, self._audio_id,
audio_slice_id), r.channels,
r.samplerate) as w:
phasevocoder(r.channels, speed=target_bpm / bar_bpm).run(r, w)
print("only came " + audio_slice_id)
def getStretchedData(low, sf):
s = "placeholder.wav"
playSpeed = 1/sf
if low:
s = "lowPlaceholder.wav"
playSpeed *= LOW_FACTOR
with WavReader(s) as reader:
with WavWriter("stretchholder.wav", reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=playSpeed)
tsm.run(reader, writer)
_, s = wavfile.read("stretchholder.wav")
d = np.zeros(s.shape)
if low:
d += s
else:
d += s*0.81
return d
def getStretchedData(low, sf):
s = PLACEHOLDER_WAV_AUX
playSpeed = 1 / sf
if low:
s = LOW_PLACEHOLDER_WAV_AUX
playSpeed *= LOW_FACTOR
with WavReader(s) as reader:
with WavWriter(STRECH, reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=playSpeed)
tsm.run(reader, writer)
_, s = wavfile.read(STRECH)
d = np.zeros(s.shape)
if low:
d += s
else:
d += s * 0.81
return d
def execute(self):
# get values of audio frames, 0 for silence, 1 for loudness.
has_loud_audio = self.get_loud_frame()
# get edit points of silence and loudness.
edit_points = self.get_edit_points(has_loud_audio)
start_frame = 0
output = self.get_output()
for edit_point in edit_points:
audio_chunk = self.parameter.audio_data[
int(edit_point.start_frame * self.parameter.samples_per_frame):
int(edit_point.end_frame * self.parameter.samples_per_frame)
]
# need channels * frames, transpose data first.
reader = ArrayReader(np.transpose(audio_chunk))
writer = ArrayWriter(reader.channels)
tsm = phasevocoder(reader.channels, speed=self.parameter.new_speed[int(edit_point.should_keep)])
tsm.run(reader, writer)
altered_audio_data = np.transpose(writer.data)
altered_audio_data_length = altered_audio_data.shape[0]
if altered_audio_data_length < self.parameter.audio_fade_envelope_size:
altered_audio_data[:] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
self.fade_out_silence(altered_audio_data)
end_frame = start_frame + altered_audio_data_length
start_output_frame = int(math.ceil(start_frame / self.parameter.samples_per_frame))
end_output_frame = int(math.ceil(end_frame / self.parameter.samples_per_frame))
output.apply_edit_point(edit_point, altered_audio_data, start_output_frame, end_output_frame)
start_frame = end_frame
output.close()
def fast_video_function(videoFile, NEW_SPEEDfloat, silentThreshold, frameMargin):
global NEW_SPEED
NEW_SPEED = [NEW_SPEEDfloat, 1]
global startTime
startTime = time.time()
global cap
cap = cv2.VideoCapture(videoFile)
#In case files were left behind
try:
os.remove('output.wav')
os.remove('spedup.mp4')
os.remove('spedupAudio.wav')
except:
pass
global width
global height
global fourcc
global fps
global extractAudio
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
fps = round(cap.get(cv2.CAP_PROP_FPS))
extractAudio = 'ffmpeg -i "{}" -ab 160k -ac 2 -ar 44100 -vn output.wav'.format(videoFile)
subprocess.call(extractAudio, shell=True)
global out
out = cv2.VideoWriter('spedup.mp4', fourcc, fps, (width, height))
sampleRate, audioData = wavfile.read('output.wav')
global skipped
skipped = 0
global nFrames
nFrames = 0
global channels
channels = int(audioData.shape[1])
framesProcessed = 0
def getMaxVolume(s):
maxv = np.max(s)
minv = np.min(s)
return max(maxv,-minv)
def writeFrames(frames, nAudio, speed, samplePerSecond, writer):
numAudioChunks = round(nAudio / samplePerSecond * fps)
global nFrames
numWrites = numAudioChunks - nFrames
# a = [1, 2, 3], len(a) == 3 but a[3] is error
limit = len(frames) - 1
for i in range(numWrites):
frameIndex = round(i * speed)
if frameIndex > limit:
writer.write(frames[-1])
else:
writer.write(frames[frameIndex])
nFrames += 1
global normal
normal = 0
# 0 for silent, 1 for normal
global switchStart
switchStart = 0
global maxVolume
maxVolume = getMaxVolume(audioData)
# not used:
# fadeInSamples = 400
# preMask = np.arange(fadeInSamples)/fadeInSamples
# mask = np.repeat(preMask[:, np.newaxis], 2, axis = 1)
global y
global yPointer
global frameBuffer
y = np.zeros_like(audioData, dtype=np.int16)
yPointer = 0
frameBuffer = []
while (cap.isOpened()):
ret, frame = cap.read()
if not ret:
break
# since samplerate is in seconds, I need to convert this to second as well
currentTime = cap.get(cv2.CAP_PROP_POS_MSEC) / 1000
audioSampleStart = math.floor(currentTime * sampleRate)
# more accurate frame counting
framesProcessed += 1
# audioSampleStart + one frame worth of samples
audioSampleEnd = min((audioSampleStart + ((sampleRate // fps) * frameMargin)),(len(audioData)))
switchEnd = (audioSampleStart + ((sampleRate // fps)))
audioChunkMod = audioData[audioSampleStart:switchEnd]
audioChunk = audioData[audioSampleStart:audioSampleEnd]
# if it's quite
if getMaxVolume(audioChunk) / maxVolume < silentThreshold:
skipped += 1
# if the frame is 'switched'
frameBuffer.append(frame)
normal = 0
else: # if it's 'loud'
# and the last frame is 'loud'
if normal:
out.write(frame)
nFrames += 1
switchStart = switchEnd
yPointerEnd = yPointer + audioChunkMod.shape[0]
y[yPointer : yPointerEnd] = audioChunkMod
yPointer = yPointerEnd
else:
spedChunk = audioData[switchStart:switchEnd]
spedupAudio = np.zeros((0,2), dtype=np.int16)
with ArrReader(spedChunk, channels, sampleRate, 2) as reader:
with ArrWriter(spedupAudio, channels, sampleRate, 2) as writer:
tsm = phasevocoder(reader.channels, speed=NEW_SPEED[normal])
tsm.run(reader, writer)
spedupAudio = writer.output
yPointerEnd = yPointer + spedupAudio.shape[0]
y[yPointer : yPointerEnd] = spedupAudio
yPointer = yPointerEnd
writeFrames(frameBuffer, yPointerEnd, NEW_SPEED[normal], sampleRate, out)
frameBuffer = []
switchStart = switchEnd
normal = 1
if framesProcessed % 500 == 0:
print("{} frames processed".format(framesProcessed))
print("{} frames skipped".format(skipped))
y = y[:yPointer]
wavfile.write("spedupAudio.wav", sampleRate, y)
cap.release()
out.release()
cv2.destroyAllWindows()
outFile = "{}_faster{}".format(videoFile[:videoFile.rfind('.')],videoFile[videoFile.rfind('.'):])
command = "ffmpeg -y -i spedup.mp4 -i spedupAudio.wav -c:v copy -c:a aac {}".format(outFile)
subprocess.call(command, shell=True)
os.remove('output.wav')
os.remove('spedup.mp4')
os.remove('spedupAudio.wav')
timeLength = round(time.time() - startTime, 2)
minutes = timedelta(seconds=(round(timeLength)))
print('Finished.')
print(f'Took {timeLength} seconds ({minutes})')
print(f'Removed {math.floor(skipped / fps)} seconds from a {math.floor(framesProcessed / fps)} second video.')
def splitVideo(chunks, NEW_SPEED, frameRate, zooms, samplesPerFrame,
SAMPLE_RATE, audioData, extension, VERBOSE):
"""
This function is responsible for outputting a new image sequence in the correct order.
splitVideo is also responsible for creating Renames.txt. copying every jpeg is
computationally expensive, renaming the file is less so, but we still need to create
the cache folder and we can't do that if the program has no idea what it renamed
and where. Renames.txt will be read in originalMethod.py to recreate the original
image sequence.
To avoid desyncing issues with the audio, we need to have audioData and go along
roughly the same way originalAudio.py does it. Specifically, get the length of the
new audio chunk.
If the speed set is 1, this is easy. If not, we have to create a new file modified
to be the new speed with audiotsm, then read that file to get the length.
"""
print('Creating new video.')
num = 0
chunk_len = str(len(chunks))
outputPointer = 0
Renames = []
lastExisting = None
for chunk in chunks:
if (NEW_SPEED[int(chunk[2])] < 99999):
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
if (NEW_SPEED[chunk[2]] == 1):
leng = len(audioChunk)
else:
sFile = TEMP + '/tempStart2.wav'
eFile = TEMP + '/tempEnd2.wav'
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
phasevocoder(reader.channels,
speed=NEW_SPEED[chunk[2]]).run(
reader, writer)
__, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
startOutputFrame = int(math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer / samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0] + NEW_SPEED[chunk[2]] *
(outputFrame - startOutputFrame))
src = ''.join(
[CACHE, '/frame{:06d}'.format(inputFrame + 1), '.jpg'])
dst = ''.join(
[TEMP, '/newFrame{:06d}'.format(outputFrame + 1), '.jpg'])
if (os.path.isfile(src)):
lastExisting = inputFrame
if (inputFrame in zooms):
resize(src, dst, zooms[inputFrame])
else:
os.rename(src, dst)
Renames.extend([src, dst])
else:
if (lastExisting == None):
print(src + ' does not exist.')
raise IOError(f'Fatal Error! No existing frame exist.')
src = ''.join([
CACHE, '/frame{:06d}'.format(lastExisting + 1), '.jpg'
])
if (os.path.isfile(src)):
if (lastExisting in zooms):
resize(src, dst, zooms[lastExisting])
else:
os.rename(src, dst)
Renames.extend([src, dst])
else:
# uh oh, we need to find the file we just renamed!
myFile = None
for i in range(0, len(Renames), 2):
if (Renames[i] == src):
myFile = Renames[i + 1]
break
if (myFile is not None):
copyfile(myFile, dst)
else:
raise IOError(
f'Error! The file {src} does not exist.')
outputPointer = endPointer
num += 1
if (num % 10 == 0):
print(''.join([str(num), '/', chunk_len, ' frame chunks done.']))
print(''.join([str(num), '/', chunk_len, ' frame chunks done.']))
with open(f'{TEMP}/Renames.txt', 'w') as f:
for item in Renames:
f.write(f"{item}\n")
print('Creating finished video. (This can take a while)')
cmd = [
'ffmpeg', '-y', '-framerate',
str(frameRate), '-i', f'{TEMP}/newFrame%06d.jpg',
f'{TEMP}/output{extension}'
]
if (not VERBOSE):
cmd.extend(['-nostats', '-loglevel', '0'])
subprocess.call(cmd)
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
sFile = TEMP_FOLDER + "/tempStart.wav"
eFile = TEMP_FOLDER + "/tempEnd.wav"
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
#outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
def process(self):
global isCanceld
try:
if os.path.exists(TEMP_FOLDER):
msg = input(
'Warning: Are you sure you want to Delete videocuts_tmp folder'
)
if msg.lower() in ('yes', 'y'):
deletePath(TEMP_FOLDER)
Thread(target=self.timer).start()
global gain
gain = 1.2
self.new_video_size = 'N/A'
self.new_video_length = 'N/A'
Extras = ""
frameRate = float(60)
SAMPLE_RATE = int(self.frame_rate)
SILENT_THRESHOLD = float(self.silence_threshold)
FRAME_SPREADAGE = int(self.frame_margin)
NEW_SPEED = [float(self.silent_speed), float(self.play_speed)]
gain = 0.6
INPUT_FILE = self.downloadFile(str(self.video_url))
if INPUT_FILE == '':
return
FRAME_QUALITY = self.frame_quality
assert INPUT_FILE is not None, "You did not specify an input file. You must specify an input file without spaces."
OUTPUT_FILE = inputToOutputFilename(INPUT_FILE)
AUDIO_FADE_ENVELOPE_SIZE = 400 # smooth out transitiion's audio by quickly fading in/out (arbitrary magic number whatever)
createPath(TEMP_FOLDER)
global dir
dir = os.getcwd()
if isCanceld:
return
print(
' Step 1 - Frame quality has been assessed and is processing ')
cmdary = [
resource_path('ffmpeg.exe'), "-i", INPUT_FILE, '-qscale:v',
str(FRAME_QUALITY), TEMP_FOLDER + "/frame%06d.jpg",
'-hide_banner'
]
subprocess.call(cmdary, cwd=dir, shell=True)
if isCanceld:
return
print(
' Step 1 - Frame quality processing has successfully completed '
)
time.sleep(2)
if isCanceld:
return
print(' Step 2 - Sample Rate has been assessed and is processing ')
cmdary = [
resource_path('ffmpeg.exe'), "-i", INPUT_FILE, '2>&1', '-ab',
'160k', '-ac', '2', '-ar',
str(SAMPLE_RATE), '-vn', TEMP_FOLDER + "/audio.wav"
]
subprocess.call(cmdary, cwd=dir, shell=True)
if isCanceld:
return
print(
' Step 2 - Sample Rate processing has successfully completed ')
time.sleep(2)
if isCanceld:
return
print(
' Step 3 - Video Frames are processing. This might take a while... '
)
cmdary = [resource_path('ffmpeg.exe'), "-i", INPUT_FILE, '2>&1']
open(TEMP_FOLDER + "/params.txt", "w")
subprocess.call(cmdary, cwd=dir, shell=True)
if isCanceld:
return
print(
' Step 3 - Video Frames processing has successfully completed '
)
time.sleep(2)
sampleRate, audioData = wavfile.read(TEMP_FOLDER + "/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = getMaxVolume(audioData)
cap = cv2.VideoCapture(INPUT_FILE)
fps = cap.get(cv2.CAP_PROP_FPS)
f = open(TEMP_FOLDER + "/params.txt", 'r+')
pre_params = f.read()
f.close()
params = pre_params.split('\n')
for line in params:
m = re.search(' ([0-9]*.[0-9]*) fps,', line)
if m is None:
frameRate = float(fps)
if m is not None:
frameRate = float(m.group(1))
samplesPerFrame = sampleRate / frameRate
audioFrameCount = int(math.ceil(audioSampleCount /
samplesPerFrame))
hasLoudAudio = np.zeros(audioFrameCount)
for i in range(audioFrameCount):
start = int(i * samplesPerFrame)
end = min(int((i + 1) * samplesPerFrame), audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(
getMaxVolume(audiochunks)) / maxAudioVolume
if maxchunksVolume >= SILENT_THRESHOLD:
hasLoudAudio[i] = 1
chunks = [[0, 0, 0]]
shouldIncludeFrame = np.zeros(audioFrameCount)
for i in range(audioFrameCount):
start = int(max(0, i - FRAME_SPREADAGE))
end = int(min(audioFrameCount, i + 1 + FRAME_SPREADAGE))
shouldIncludeFrame[i] = np.max(hasLoudAudio[start:end])
if i >= 1 and shouldIncludeFrame[i] != shouldIncludeFrame[
i - 1]: # Did we flip?
chunks.append(
[chunks[-1][1], i, shouldIncludeFrame[i - 1]])
chunks.append(
[chunks[-1][1], audioFrameCount, shouldIncludeFrame[i - 1]])
chunks = chunks[1:]
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] * samplesPerFrame
):int(chunk[1] * samplesPerFrame)]
sFile = TEMP_FOLDER + "/tempStart.wav"
eFile = TEMP_FOLDER + "/tempEnd.wav"
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
# outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(
AUDIO_FADE_ENVELOPE_SIZE) / AUDIO_FADE_ENVELOPE_SIZE
mask = np.repeat(
premask[:, np.newaxis], 2,
axis=1) # make the fade-envelope mask stereo
outputAudioData[outputPointer:outputPointer +
AUDIO_FADE_ENVELOPE_SIZE] *= mask
outputAudioData[
endPointer -
AUDIO_FADE_ENVELOPE_SIZE:endPointer] *= 1 - mask
startOutputFrame = int(
math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer / samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0] + NEW_SPEED[int(chunk[2])] *
(outputFrame - startOutputFrame))
didItWork = copyFrame(inputFrame, outputFrame)
if didItWork:
lastExistingFrame = inputFrame
else:
copyFrame(lastExistingFrame, outputFrame)
outputPointer = endPointer
wavfile.write(TEMP_FOLDER + "/audioNew.wav", SAMPLE_RATE,
outputAudioData)
'''
outputFrame = math.ceil(outputPointer/samplesPerFrame)
for endGap in range(outputFrame,audioFrameCount):
copyFrame(int(audioSampleCount/samplesPerFrame)-1,endGap)
'''
if isCanceld:
return
print(' Step 4 - Finalizing.... Please wait')
cmdary = [
resource_path('ffmpeg.exe'), '-framerate',
str(frameRate), "-i", TEMP_FOLDER + "/newFrame%06d.jpg", '-i',
TEMP_FOLDER + "/audioNew.wav", '-strict', '-2' + str(Extras),
OUTPUT_FILE
]
subprocess.call(cmdary, cwd=dir, shell=True)
if isCanceld:
return
print(' Video processing finished successfully.')
deletePath(TEMP_FOLDER)
path = os.path.dirname(INPUT_FILE)
global stpTimer
stpTimer = True
self.new_video_size = get_size(OUTPUT_FILE) + ' MB'
self.output_parameters.append(self.new_video_size)
self.new_video_length = str(self.get_length(OUTPUT_FILE))
self.output_parameters.append(self.new_video_length)
except Exception as e:
print(' Processing Video Failed! ')
if str(e) != 'main thread is not in main loop':
print('error message.', str(e))
deletePath(TEMP_FOLDER)
print(self.output_parameters)
def run_process():
URL = self.URLLineEdit.text()
if (self.URLRadio.isChecked()):
INPUT_FILE = downloadFile(URL)
else:
INPUT_FILE = self.fileLineEdit.text()
if (INPUT_FILE == ''):
winsound.PlaySound('SystemExclamation', winsound.SND_ALIAS)
else:
frameRate = self.frameRateSlider.value()
SAMPLE_RATE = self.sampleRateSlider.value()
SILENT_THRESHOLD = float(self.thresholdLineEdit.text())
FRAME_SPREADAGE = float(self.frameMarginSlider.value())
SILENT_SPEED = float(self.silentSpeedLineEdit.text())
SOUNDED_SPEED = float(self.soundedSpeedLineEdit.text())
NEW_SPEED = [SILENT_SPEED, SOUNDED_SPEED]
print(NEW_SPEED)
FRAME_QUALITY = float(self.frameQualitySlider.value())
assert INPUT_FILE != None, "why u put no input file, that dum"
"""
if len(args.output_file) >= 1:
OUTPUT_FILE = args.output_file
else:
"""
OUTPUT_FILE = inputToOutputFilename(INPUT_FILE)
TEMP_FOLDER = "TEMP"
AUDIO_FADE_ENVELOPE_SIZE = 400 # smooth out transitiion's audio by quickly fading in/out (arbitrary magic number whatever)
createPath(TEMP_FOLDER)
command = "ffmpeg -i " + INPUT_FILE + " -qscale:v " + str(
FRAME_QUALITY
) + " " + TEMP_FOLDER + "/frame%06d.jpg -hide_banner"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + INPUT_FILE + " -ab 160k -ac 2 -ar " + str(
SAMPLE_RATE) + " -vn " + TEMP_FOLDER + "/audio.wav"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + TEMP_FOLDER + "/input.mp4 2>&1"
f = open(TEMP_FOLDER + "/params.txt", "w")
subprocess.call(command, shell=True, stdout=f)
sampleRate, audioData = wavfile.read(TEMP_FOLDER +
"/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = getMaxVolume(audioData)
f = open(TEMP_FOLDER + "/params.txt", 'r+')
pre_params = f.read()
f.close()
params = pre_params.split('\n')
for line in params:
m = re.search('Stream #.*Video.* ([0-9]*) fps', line)
if m is not None:
frameRate = float(m.group(1))
samplesPerFrame = sampleRate / frameRate
audioFrameCount = int(
math.ceil(audioSampleCount / samplesPerFrame))
hasLoudAudio = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(i * samplesPerFrame)
end = min(int((i + 1) * samplesPerFrame), audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(
getMaxVolume(audiochunks)) / maxAudioVolume
if maxchunksVolume >= SILENT_THRESHOLD:
hasLoudAudio[i] = 1
chunks = [[0, 0, 0]]
shouldIncludeFrame = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(max(0, i - FRAME_SPREADAGE))
end = int(min(audioFrameCount, i + 1 + FRAME_SPREADAGE))
shouldIncludeFrame[i] = np.max(hasLoudAudio[start:end])
if (i >= 1 and shouldIncludeFrame[i] !=
shouldIncludeFrame[i - 1]): # Did we flip?
chunks.append(
[chunks[-1][1], i, shouldIncludeFrame[i - 1]])
chunks.append([
chunks[-1][1], audioFrameCount, shouldIncludeFrame[i - 1]
])
chunks = chunks[1:]
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] * samplesPerFrame
):int(chunk[1] *
samplesPerFrame)]
sFile = TEMP_FOLDER + "/tempStart.wav"
eFile = TEMP_FOLDER + "/tempEnd.wav"
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
#outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(AUDIO_FADE_ENVELOPE_SIZE
) / AUDIO_FADE_ENVELOPE_SIZE
mask = np.repeat(
premask[:, np.newaxis], 2,
axis=1) # make the fade-envelope mask stereo
outputAudioData[outputPointer:outputPointer +
AUDIO_FADE_ENVELOPE_SIZE] *= mask
outputAudioData[
endPointer -
AUDIO_FADE_ENVELOPE_SIZE:endPointer] *= 1 - mask
startOutputFrame = int(
math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(
math.ceil(endPointer / samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0] + NEW_SPEED[int(chunk[2])] *
(outputFrame - startOutputFrame))
didItWork = copyFrame(inputFrame, outputFrame,
TEMP_FOLDER)
if didItWork:
lastExistingFrame = inputFrame
else:
copyFrame(lastExistingFrame, outputFrame,
TEMP_FOLDER)
outputPointer = endPointer
wavfile.write(TEMP_FOLDER + "/audioNew.wav", SAMPLE_RATE,
outputAudioData)
'''
outputFrame = math.ceil(outputPointer/samplesPerFrame)
for endGap in range(outputFrame,audioFrameCount):
copyFrame(int(audioSampleCount/samplesPerFrame)-1,endGap)
'''
command = "ffmpeg -framerate " + str(
frameRate
) + " -i " + TEMP_FOLDER + "/newFrame%06d.jpg -i " + TEMP_FOLDER + "/audioNew.wav -strict -2 " + OUTPUT_FILE
subprocess.call(command, shell=True)
deletePath(TEMP_FOLDER)
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
sFile = os.path.join(TEMP_FOLDER, "tempStart.wav")
eFile = os.path.join(TEMP_FOLDER, "tempEnd.wav")
outputFrame = 0
for chunk in chunks:
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
phasevocoder(reader.channels, speed=NEW_SPEED[int(chunk[2])]).run(
reader, WavWriter(eFile, reader.channels, reader.samplerate))
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
# outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transition's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
signalborders = np.where(signalpos_smoothed[:np.size(signalpos_smoothed)-1] != signalpos_smoothed[1:])[0]
del signalpos_smoothed
#signalpos = np.where(np.abs(content) > threschold)
#signalpos_norepeat = np.unique(signalpos[0])
#signalpos_norepeat = np.append(signalpos_norepeat, np.shape(content)[0])
#signalborders = signalpos_norepeat[np.gradient(signalpos_norepeat) > 2000]
signalborders = np.insert(signalborders, 0, 0)
signalborders = np.append(signalborders, np.size(content[:,0]))
newcontent = np.empty((0,2), dtype=np.int16)
for i in (np.arange(1, np.size(signalborders))):
if np.mean(np.abs(content[signalborders[i-1]:signalborders[i],:])) > threschold:
lborder = int(np.max([signalborders[i-1]-rate/15, 0]))
uborder = int(np.min([signalborders[i]+rate/15, np.size(content[:,0])]))
acc_size = int(np.floor((uborder-lborder)/acc_rate0))
acc_part = np.empty((acc_size,2))
nonacc_part = content[lborder:uborder,:]
acc_part = nonacc_part[np.floor(np.arange(acc_size) * acc_rate0).astype(int),:]
newcontent = np.append(newcontent, acc_part, axis=0)
wf.write('output_temp.wav', rate, newcontent)
with WavReader('output_temp.wav') as reader:
with WavWriter('output.wav', reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=acc_rate)
tsm.run(reader, writer)
os.remove('output_temp.wav')
def splitAudio(filename, chunks, samplesPerFrame, NEW_SPEED, audioData,
SAMPLE_RATE, maxAudioVolume):
"""
This function creates new audio based on the chunk date and the numpy audio data.
"""
outputAudioData = []
outputPointer = 0
mask = [x / FADE_SIZE for x in range(FADE_SIZE)]
num = 0
chunk_len = str(len(chunks))
for chunk in chunks:
if (NEW_SPEED[chunk[2]] < 99999):
start = int(chunk[0] * samplesPerFrame)
end = int(chunk[1] * samplesPerFrame)
audioChunk = audioData[start:end]
sFile = ''.join([TEMP, '/tempStart.wav'])
eFile = ''.join([TEMP, '/tempEnd.wav'])
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
if (NEW_SPEED[chunk[2]] == 1):
__, samefile = wavfile.read(sFile)
leng = len(audioChunk)
outputAudioData.extend((samefile / maxAudioVolume).tolist())
else:
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
phasevocoder(reader.channels,
speed=NEW_SPEED[chunk[2]]).run(
reader, writer)
__, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
outputAudioData.extend(
(alteredAudioData / maxAudioVolume).tolist())
endPointer = outputPointer + leng
# smooth out transition's audio by quickly fading in/out
if (leng < FADE_SIZE):
for i in range(outputPointer, endPointer):
try:
outputAudioData[i][0] = 0
outputAudioData[i][1] = 0
except TypeError:
outputAudioData[i] = 0
else:
for i in range(outputPointer, outputPointer + FADE_SIZE):
try:
outputAudioData[i][0] *= mask[i - outputPointer]
outputAudioData[i][1] *= mask[i - outputPointer]
except TypeError:
outputAudioData[i] *= mask[i - outputPointer]
for i in range(endPointer - FADE_SIZE, endPointer):
try:
outputAudioData[i][0] *= (
1 - mask[i - endPointer + FADE_SIZE])
outputAudioData[i][1] *= (
1 - mask[i - endPointer + FADE_SIZE])
except TypeError:
outputAudioData[i] *= (
1 - mask[i - endPointer + FADE_SIZE])
outputPointer = endPointer
num += 1
if (num % 10 == 0):
print(''.join([str(num), '/', chunk_len, ' audio chunks done.']))
print(''.join([str(num), '/', chunk_len, ' audio chunks done.']))
outputAudioData = np.asarray(outputAudioData)
wavfile.write(filename, SAMPLE_RATE, outputAudioData)
if (not os.path.isfile(filename)):
raise IOError(f'Error: The file {filename} was not created.')
else:
print('Audio finished.')
def execute(input_file="",
url="",
output_file="",
silent_threshold=0.03,
sounded_speed=1.00,
silent_speed=5.00,
frame_margin=1,
sample_rate=44100,
frame_quality=3):
SAMPLE_RATE = sample_rate
SILENT_THRESHOLD = silent_threshold
FRAME_SPREADAGE = frame_margin
NEW_SPEED = [silent_speed, sounded_speed]
if url != "" and url != None:
INPUT_FILE = downloadFile(url)
else:
INPUT_FILE = input_file
FRAME_QUALITY = frame_quality
assert INPUT_FILE != "" and INPUT_FILE != None, "why u put no input file, that dum"
if len(output_file) >= 1:
OUTPUT_FILE = output_file
else:
OUTPUT_FILE = inputToOutputFilename(INPUT_FILE)
print("Saving to: " + OUTPUT_FILE)
AUDIO_FADE_ENVELOPE_SIZE = 400 # smooth out transitiion's audio by quickly fading in/out (arbitrary magic number whatever)
createPath(TEMP_FOLDER)
checkForFFMPEG()
command = "ffmpeg -i " + INPUT_FILE + " -qscale:v " + str(
FRAME_QUALITY) + " " + TEMP_FOLDER + "/frame%06d.jpg -hide_banner"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + INPUT_FILE + " -ab 160k -ac 2 -ar " + str(
SAMPLE_RATE) + " -vn " + TEMP_FOLDER + "/audio.wav"
subprocess.call(command, shell=True)
sampleRate, audioData = wavfile.read(TEMP_FOLDER + "/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = getMaxVolume(audioData)
frameRate = findFramerate(INPUT_FILE)
samplesPerFrame = sampleRate / frameRate
audioFrameCount = int(math.ceil(audioSampleCount / samplesPerFrame))
hasLoudAudio = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(i * samplesPerFrame)
end = min(int((i + 1) * samplesPerFrame), audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(getMaxVolume(audiochunks)) / maxAudioVolume
if maxchunksVolume >= SILENT_THRESHOLD:
hasLoudAudio[i] = 1
chunks = [[0, 0, 0]]
shouldIncludeFrame = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(max(0, i - FRAME_SPREADAGE))
end = int(min(audioFrameCount, i + 1 + FRAME_SPREADAGE))
shouldIncludeFrame[i] = np.max(hasLoudAudio[start:end])
if (i >= 1 and shouldIncludeFrame[i] !=
shouldIncludeFrame[i - 1]): # Did we flip?
chunks.append([chunks[-1][1], i, shouldIncludeFrame[i - 1]])
chunks.append([chunks[-1][1], audioFrameCount, shouldIncludeFrame[i - 1]])
chunks = chunks[1:]
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
sFile = TEMP_FOLDER + "/tempStart.wav"
eFile = TEMP_FOLDER + "/tempEnd.wav"
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
# outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(
AUDIO_FADE_ENVELOPE_SIZE) / AUDIO_FADE_ENVELOPE_SIZE
mask = np.repeat(premask[:, np.newaxis], 2,
axis=1) # make the fade-envelope mask stereo
outputAudioData[outputPointer:outputPointer +
AUDIO_FADE_ENVELOPE_SIZE] *= mask
outputAudioData[endPointer -
AUDIO_FADE_ENVELOPE_SIZE:endPointer] *= 1 - mask
startOutputFrame = int(math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer / samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0] + NEW_SPEED[int(chunk[2])] *
(outputFrame - startOutputFrame))
didItWork = copyFrame(inputFrame, outputFrame)
if didItWork:
lastExistingFrame = inputFrame
else:
copyFrame(lastExistingFrame, outputFrame)
outputPointer = endPointer
wavfile.write(TEMP_FOLDER + "/audioNew.wav", SAMPLE_RATE, outputAudioData)
'''
outputFrame = math.ceil(outputPointer/samplesPerFrame)
for endGap in range(outputFrame,audioFrameCount):
copyFrame(int(audioSampleCount/samplesPerFrame)-1,endGap)
'''
command = "ffmpeg -y -framerate " + str(
frameRate
) + " -i " + TEMP_FOLDER + "/newFrame%06d.jpg -i " + TEMP_FOLDER + "/audioNew.wav -strict -2 " + OUTPUT_FILE
subprocess.call(command, shell=True)
deletePath(TEMP_FOLDER)
def process_video(args):
TEMP_FOLDER,frameRate,SAMPLE_RATE,NEW_SPEED,SILENT_THRESHOLD,FRAME_SPREADAGE,AUDIO_FADE_ENVELOPE_SIZE = args
sampleRate, audioData = wavfile.read(TEMP_FOLDER+"/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = getMaxVolume(audioData)
samplesPerFrame = sampleRate/frameRate
audioFrameCount = int(math.ceil(audioSampleCount/samplesPerFrame))
hasLoudAudio = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(i*samplesPerFrame)
end = min(int((i+1)*samplesPerFrame),audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(getMaxVolume(audiochunks))/maxAudioVolume
if maxchunksVolume >= SILENT_THRESHOLD:
hasLoudAudio[i] = 1
chunks = [[0,0,0]]
shouldIncludeFrame = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(max(0,i-FRAME_SPREADAGE))
end = int(min(audioFrameCount,i+1+FRAME_SPREADAGE))
shouldIncludeFrame[i] = np.max(hasLoudAudio[start:end])
if (i >= 1 and shouldIncludeFrame[i] != shouldIncludeFrame[i-1]): # Did we flip?
chunks.append([chunks[-1][1],i,shouldIncludeFrame[i-1]])
chunks.append([chunks[-1][1],audioFrameCount,shouldIncludeFrame[i-1]])
chunks = chunks[1:]
outputAudioData = np.zeros((0,audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0]*samplesPerFrame):int(chunk[1]*samplesPerFrame)]
sFile = TEMP_FOLDER+"/tempStart.wav"
eFile = TEMP_FOLDER+"/tempEnd.wav"
wavfile.write(sFile,SAMPLE_RATE,audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer+leng
outputAudioData = np.concatenate((outputAudioData,alteredAudioData/maxAudioVolume))
#outputAudioData[outputPointer:endPointer] = alteredAudioData/maxAudioVolume
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[outputPointer:endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(AUDIO_FADE_ENVELOPE_SIZE)/AUDIO_FADE_ENVELOPE_SIZE
mask = np.repeat(premask[:, np.newaxis],2,axis=1) # make the fade-envelope mask stereo
outputAudioData[outputPointer:outputPointer+AUDIO_FADE_ENVELOPE_SIZE] *= mask
outputAudioData[endPointer-AUDIO_FADE_ENVELOPE_SIZE:endPointer] *= 1-mask
startOutputFrame = int(math.ceil(outputPointer/samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer/samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0]+NEW_SPEED[int(chunk[2])]*(outputFrame-startOutputFrame))
didItWork = copyFrame(TEMP_FOLDER,inputFrame,outputFrame)
if didItWork:
lastExistingFrame = inputFrame
else:
copyFrame(TEMP_FOLDER,lastExistingFrame,outputFrame)
outputPointer = endPointer
wavfile.write(TEMP_FOLDER+"/audioNew.wav",SAMPLE_RATE,outputAudioData)
'''
if normal:
out.write(frame)
nFrames += 1
switchStart = switchEnd
yPointerEnd = yPointer + audioChunk.shape[0]
y[yPointer:yPointerEnd] = audioChunk
yPointer = yPointerEnd
else:
spedChunk = audioData[switchStart:switchEnd]
spedupAudio = np.zeros((0, 2), dtype=np.int16)
# ArrReader (array, channels, samplerate, samplewidth)
with ArrReader(spedChunk, channels, sampleRate, 2) as reader:
# 2 as sampleWidth for now
with ArrWriter(spedupAudio, channels, sampleRate, 2) as writer:
tsm = phasevocoder(reader.channels, speed=silentSpeed)
tsm.run(reader, writer)
spedupAudio = writer.output
yPointerEnd = yPointer + spedupAudio.shape[0]
y[yPointer:yPointerEnd] = spedupAudio
yPointer = yPointerEnd
writeFrames(frameBuffer, yPointerEnd, silentSpeed, sampleRate, out)
frameBuffer = []
switchStart = switchEnd
normal = 1
if skipped % 1000 == 0:
print("{} frames inspected".format(skipped))
skipped += 1
def process_and_concatenate(self):
audio_fade_envelope_size = 400 # smooth out transition's audio by quickly fading in/out
self.save_audio()
sample_rate, audio_data = wavfile.read(self.temp_folder + "/audio.wav")
audio_sample_count = audio_data.shape[0]
max_audio_volume = get_max_volume(audio_data)
samples_per_frame = sample_rate / self.fps
audio_frame_count = int(
math.ceil(audio_sample_count / samples_per_frame))
has_loud_audio = np.zeros(audio_frame_count)
for i in range(audio_frame_count):
start = int(i * samples_per_frame)
end = min(int((i + 1) * samples_per_frame), audio_sample_count)
audio_chunks = audio_data[start:end]
max_chunks_volume = float(
get_max_volume(audio_chunks)) / max_audio_volume
if max_chunks_volume >= SILENT_THRESHOLD:
has_loud_audio[i] = 1
chunks = [[0, 0, 0]]
should_include_frame = np.zeros(audio_frame_count)
last_idx = 0
for i in range(audio_frame_count):
start = int(max(0, i - FRAME_SPREADAGE))
end = int(min(audio_frame_count, i + 1 + FRAME_SPREADAGE))
should_include_frame[i] = np.max(has_loud_audio[start:end])
if i >= 1 and should_include_frame[i] != should_include_frame[
i - 1]: # Did we flip?
chunks.append([chunks[-1][1], i, should_include_frame[i - 1]])
last_idx = i
chunks.append([
chunks[-1][1], audio_frame_count,
should_include_frame[last_idx - 1]
])
chunks = chunks[1:]
output_audio_data = np.zeros((0, audio_data.shape[1]))
output_pointer = 0
last_existing_frame = None
duration = self.get_duration()
frames_num = int(float(duration) * self.fps)
signed_frames = [False for _ in range(frames_num)]
output_frames = []
for chunk in chunks:
audio_chunk = audio_data[int(chunk[0] * samples_per_frame
):int(chunk[1] * samples_per_frame)]
s_file = self.temp_folder + "/tempStart.wav"
e_file = self.temp_folder + "/tempEnd.wav"
wavfile.write(s_file, SAMPLE_RATE, audio_chunk)
with WavReader(s_file) as reader:
with WavWriter(e_file, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, altered_audio_data = wavfile.read(e_file)
leng = altered_audio_data.shape[0]
end_pointer = output_pointer + leng
output_audio_data = np.concatenate(
(output_audio_data, altered_audio_data / max_audio_volume))
if leng < audio_fade_envelope_size:
output_audio_data[output_pointer:end_pointer] = 0
else:
pre_mask = np.arange(
audio_fade_envelope_size) / audio_fade_envelope_size
mask = np.repeat(pre_mask[:, np.newaxis], 2, axis=1)
output_audio_data[output_pointer:output_pointer +
audio_fade_envelope_size] *= mask
output_audio_data[
end_pointer -
audio_fade_envelope_size:end_pointer] *= 1 - mask
start_output_frame = int(
math.ceil(output_pointer / samples_per_frame))
end_output_frame = int(math.ceil(end_pointer / samples_per_frame))
for outputFrame in range(start_output_frame, end_output_frame):
input_frame = int(chunk[0] + NEW_SPEED[int(chunk[2])] *
(outputFrame - start_output_frame))
if input_frame < frames_num - 2:
signed_frames[input_frame + 1] = True
last_existing_frame = input_frame
else:
signed_frames[last_existing_frame] = True
output_frames.append(outputFrame)
output_pointer = end_pointer
j = 0
for i, frame_sign in enumerate(signed_frames):
if frame_sign:
self.copy_frame(i, j)
j += 1
wavfile.write(self.temp_folder + "/audioNew.wav", SAMPLE_RATE,
output_audio_data)
self.final_concatenation()
delete_path(self.temp_folder)
def timeStretch(input_filename, output_filename, rate, samplerate):
with WavReader(input_filename) as reader:
with WavWriter(output_filename, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, rate)
tsm.run(reader, writer)
def main(args):
frameRate = args.frame_rate
SAMPLE_RATE = args.sample_rate
SILENT_THRESHOLD = args.silent_threshold
FRAME_SPREADAGE = args.frame_margin
NEW_SPEED = [args.silent_speed, args.sounded_speed]
FRAME_QUALITY = args.frame_quality
AUDIO_FADE_ENVELOPE_SIZE = 400 # smooth out transitiion's audio by quickly fading in/out (arbitrary magic number whatever)
TEMP_FOLDER = "TEMP"
if not os.path.isdir(TEMP_FOLDER):
os.makedirs(TEMP_FOLDER)
if args.url != None:
INPUT_FILE = downloadFile(args.url)
else:
INPUT_FILE = args.input
if len(args.output) >= 1:
OUTPUT_FILE = args.output
else:
OUTPUT_FILE = inputToOutputFilename(INPUT_FILE)
command = "ffmpeg -i " + INPUT_FILE + " -qscale:v " + str(
FRAME_QUALITY) + " " + TEMP_FOLDER + "/frame%06d.jpg -hide_banner"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + INPUT_FILE + " -ab 160k -ac 2 -ar " + str(
SAMPLE_RATE) + " -vn " + TEMP_FOLDER + "/audio.wav"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + TEMP_FOLDER + "/input.mp4 2>&1"
f = open(TEMP_FOLDER + "/params.txt", "w")
subprocess.call(command, shell=True, stdout=f)
sampleRate, audioData = wavfile.read(TEMP_FOLDER + "/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = getMaxVolume(audioData)
f = open(TEMP_FOLDER + "/params.txt", 'r+')
pre_params = f.read()
f.close()
params = pre_params.split('\n')
for line in params:
m = re.search('Stream #.*Video.* ([0-9]*) fps', line)
if m is not None:
frameRate = float(m.group(1))
samplesPerFrame = sampleRate / frameRate
audioFrameCount = int(math.ceil(audioSampleCount / samplesPerFrame))
hasLoudAudio = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(i * samplesPerFrame)
end = min(int((i + 1) * samplesPerFrame), audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(getMaxVolume(audiochunks)) / maxAudioVolume
if maxchunksVolume >= SILENT_THRESHOLD:
hasLoudAudio[i] = 1
chunks = [[0, 0, 0]]
shouldIncludeFrame = np.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(max(0, i - FRAME_SPREADAGE))
end = int(min(audioFrameCount, i + 1 + FRAME_SPREADAGE))
shouldIncludeFrame[i] = np.max(hasLoudAudio[start:end])
if (i >= 1 and shouldIncludeFrame[i] !=
shouldIncludeFrame[i - 1]): # Did we flip?
chunks.append([chunks[-1][1], i, shouldIncludeFrame[i - 1]])
chunks.append([chunks[-1][1], audioFrameCount, shouldIncludeFrame[i - 1]])
chunks = chunks[1:]
outputAudioData = np.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
sFile = TEMP_FOLDER + "/tempStart.wav"
eFile = TEMP_FOLDER + "/tempEnd.wav"
wavfile.write(sFile, SAMPLE_RATE, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[int(chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = np.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
# smooth out transitiion's audio by quickly fading in/out
if leng < AUDIO_FADE_ENVELOPE_SIZE:
outputAudioData[
outputPointer:
endPointer] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(
AUDIO_FADE_ENVELOPE_SIZE) / AUDIO_FADE_ENVELOPE_SIZE
mask = np.repeat(premask[:, np.newaxis], 2,
axis=1) # make the fade-envelope mask stereo
outputAudioData[outputPointer:outputPointer +
AUDIO_FADE_ENVELOPE_SIZE] *= mask
outputAudioData[endPointer -
AUDIO_FADE_ENVELOPE_SIZE:endPointer] *= 1 - mask
startOutputFrame = int(math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer / samplesPerFrame))
for outputFrame in range(startOutputFrame, endOutputFrame):
inputFrame = int(chunk[0] + NEW_SPEED[int(chunk[2])] *
(outputFrame - startOutputFrame))
didItWork = copyFrame(TEMP_FOLDER, inputFrame, outputFrame)
if didItWork:
lastExistingFrame = inputFrame
else:
copyFrame(TEMP_FOLDER, lastExistingFrame, outputFrame)
outputPointer = endPointer
wavfile.write(TEMP_FOLDER + "/audioNew.wav", SAMPLE_RATE, outputAudioData)
command = "ffmpeg -framerate " + str(
frameRate
) + " -i " + TEMP_FOLDER + "/newFrame%06d.jpg -i " + TEMP_FOLDER + "/audioNew.wav -strict -2 " + OUTPUT_FILE
subprocess.call(command, shell=True)
rmtree(TEMP_FOLDER, ignore_errors=False)
def speed_up_video(input_file: str,
output_file: str = None,
frame_rate: float = 30,
sample_rate: int = 44100,
silent_threshold: float = 0.03,
silent_speed: float = 5.0,
sounded_speed: float = 1.0,
frame_spreadage: int = 1,
audio_fade_envelope_size: int = 400,
temp_folder: str = 'TEMP') -> None:
"""
Speeds up a video file with different speeds for the silent and loud sections in the video.
:param input_file: The file name of the video to be sped up.
:param output_file: The file name of the output file. If not given will be 'input_file'_ALTERED.ext.
:param frame_rate: The frame rate of the given video. Only needed if not extractable through ffmpeg.
:param sample_rate: The sample rate of the audio in the video.
:param silent_threshold: The threshold when a chunk counts towards being a silent chunk.
Value ranges from 0 (nothing) - 1 (max volume).
:param silent_speed: The speed of the silent chunks.
:param sounded_speed: The speed of the loud chunks.
:param frame_spreadage: How many silent frames adjacent to sounded frames should be included to provide context.
:param audio_fade_envelope_size: Audio transition smoothing duration in samples.
:param temp_folder: The file path of the temporary working folder.
"""
# Set output file name based on input file name if none was given
if output_file is None:
output_file = _input_to_output_filename(input_file)
# Create Temp Folder
if os.path.exists(temp_folder):
_delete_path(temp_folder)
_create_path(temp_folder)
# Find out framerate and duration of the input video
command = 'ffprobe -i "{}" -hide_banner -loglevel error -select_streams v' \
' -show_entries format=duration:stream=avg_frame_rate'.format(input_file)
p = subprocess.Popen(command,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
bufsize=1,
universal_newlines=True,
shell=True)
std_out, err = p.communicate()
match_frame_rate = re.search(r'frame_rate=(\d*)/(\d*)', str(std_out))
if match_frame_rate is not None:
frame_rate = float(match_frame_rate.group(1)) / float(
match_frame_rate.group(2))
# print(f'Found Framerate {frame_rate}')
match_duration = re.search(r'duration=([\d.]*)', str(std_out))
original_duration = 0.0
if match_duration is not None:
original_duration = float(match_duration.group(1))
# print(f'Found Duration {original_duration}')
# Extract the audio
command = 'ffmpeg -i "{}" -ab 160k -ac 2 -ar {} -vn {} -hide_banner' \
.format(input_file,
sample_rate,
temp_folder + '/audio.wav')
_run_timed_ffmpeg_command(command,
total=int(original_duration * frame_rate),
unit='frames',
desc='Extracting audio:')
wav_sample_rate, audio_data = wavfile.read(temp_folder + "/audio.wav")
audio_sample_count = audio_data.shape[0]
max_audio_volume = _get_max_volume(audio_data)
samples_per_frame = wav_sample_rate / frame_rate
audio_frame_count = int(math.ceil(audio_sample_count / samples_per_frame))
# Find frames with loud audio
has_loud_audio = np.zeros(audio_frame_count, dtype=bool)
for i in range(audio_frame_count):
start = int(i * samples_per_frame)
end = min(int((i + 1) * samples_per_frame), audio_sample_count)
audio_chunk = audio_data[start:end]
chunk_max_volume = float(
_get_max_volume(audio_chunk)) / max_audio_volume
if chunk_max_volume >= silent_threshold:
has_loud_audio[i] = True
# Chunk the frames together that are quiet or loud
chunks = [[0, 0, 0]]
should_include_frame = np.zeros(audio_frame_count, dtype=bool)
for i in tqdm(range(audio_frame_count),
desc='Finding chunks:',
unit='frames'):
start = int(max(0, i - frame_spreadage))
end = int(min(audio_frame_count, i + 1 + frame_spreadage))
should_include_frame[i] = np.any(has_loud_audio[start:end])
if i >= 1 and should_include_frame[i] != should_include_frame[
i - 1]: # Did we flip?
chunks.append([chunks[-1][1], i, should_include_frame[i - 1]])
chunks.append([
chunks[-1][1], audio_frame_count,
should_include_frame[audio_frame_count - 1]
])
chunks = chunks[1:]
# Generate audio data with varying speed for each chunk
new_speeds = [silent_speed, sounded_speed]
output_pointer = 0
audio_buffers = []
for index, chunk in tqdm(enumerate(chunks),
total=len(chunks),
desc='Changing audio:',
unit='chunks'):
audio_chunk = audio_data[int(chunk[0] *
samples_per_frame):int(chunk[1] *
samples_per_frame)]
reader = ArrayReader(np.transpose(audio_chunk))
writer = ArrayWriter(reader.channels)
tsm = phasevocoder(reader.channels, speed=new_speeds[int(chunk[2])])
tsm.run(reader, writer)
altered_audio_data = np.transpose(writer.data)
# smooth out transition's audio by quickly fading in/out
if altered_audio_data.shape[0] < audio_fade_envelope_size:
altered_audio_data[:] = 0 # audio is less than 0.01 sec, let's just remove it.
else:
premask = np.arange(
audio_fade_envelope_size) / audio_fade_envelope_size
mask = np.repeat(premask[:, np.newaxis], 2,
axis=1) # make the fade-envelope mask stereo
altered_audio_data[:audio_fade_envelope_size] *= mask
altered_audio_data[-audio_fade_envelope_size:] *= 1 - mask
audio_buffers.append(altered_audio_data / max_audio_volume)
end_pointer = output_pointer + altered_audio_data.shape[0]
start_output_frame = int(math.ceil(output_pointer / samples_per_frame))
end_output_frame = int(math.ceil(end_pointer / samples_per_frame))
chunks[index] = chunk[:2] + [start_output_frame, end_output_frame]
output_pointer = end_pointer
# print(chunks)
output_audio_data = np.concatenate(audio_buffers)
wavfile.write(temp_folder + "/audioNew.wav", sample_rate,
output_audio_data)
# Cut the video parts to length
expression = _get_tree_expression(chunks)
filter_graph_file = open(temp_folder + "/filterGraph.txt", 'w')
filter_graph_file.write(f'fps=fps={frame_rate},setpts=')
filter_graph_file.write(expression.replace(',', '\\,'))
filter_graph_file.close()
command = 'ffmpeg -i "{}" -i "{}" -filter_script:v "{}" -map 0 -map -0:a -map 1:a -c:a aac "{}"' \
' -loglevel warning -stats -y -hide_banner' \
.format(input_file,
temp_folder + '/audioNew.wav',
temp_folder + '/filterGraph.txt',
output_file)
_run_timed_ffmpeg_command(command,
total=chunks[-1][3],
unit='frames',
desc='Generating final:')
_delete_path(temp_folder)
def apply_speed_to_audio(audio, speed):
reader = ArrayReader(audio)
writer = ArrayWriter(2)
tsm = phasevocoder(reader.channels, speed)
tsm.run(reader, writer)
return writer.data
# and the last frame is 'loud'
if normal:
out.write(frame)
nFrames += 1
switchStart = switchEnd
yPointerEnd = yPointer + audioChunkMod.shape[0]
y[yPointer:yPointerEnd] = audioChunkMod
yPointer = yPointerEnd
else:
spedChunk = audioData[switchStart:switchEnd]
spedupAudio = np.zeros((0, 2), dtype=np.int16)
with ArrReader(spedChunk, channels, sampleRate, 2) as reader:
with ArrWriter(spedupAudio, channels, sampleRate, 2) as writer:
tsm = phasevocoder(reader.channels,
speed=NEW_SPEED[normal])
tsm.run(reader, writer)
spedupAudio = writer.output
yPointerEnd = yPointer + spedupAudio.shape[0]
y[yPointer:yPointerEnd] = spedupAudio
yPointer = yPointerEnd
writeFrames(frameBuffer, yPointerEnd, NEW_SPEED[normal],
sampleRate, out)
frameBuffer = []
switchStart = switchEnd
normal = 1
if framesProcessed % 500 == 0:
print("{} frames processed".format(framesProcessed))
def videoProcess(frame_rate, sample_rate, silent_threshold, frame_margin,
silent_speed, sounded_speed, url, input_file, output_file,
frame_quality):
try:
print(frame_rate, sample_rate, silent_threshold, frame_margin,
silent_speed, sounded_speed, url, input_file, output_file,
frame_quality)
New_Speed_silent_and_sounded = [silent_speed, sounded_speed]
if url:
name = YouTube(url).streams.first().download()
renamed = name.replace(' ', '_')
os.rename(name, renamed)
return renamed
else:
Input_Video = input_file
assert Input_Video != None, "enter input video"
if len(output_file) >= 1:
Output_Video = output_file
else:
dot_position = filename.rfind(".")
Output_Video = filename[:dot_position] + "NEWVIDEO" + filename[
dot_position:]
# print ( Output_Video)
Audio_fade_envelope_size = 400
try:
os.mkdir(TEMP)
except OSError:
assert False, "Directory Already existing"
command = "ffmpeg -i " + Input_Video + " -qscale:v " + str(
frame_quality) + " " + TEMP + "/old_frame%06d.jpg -hide_banner"
subprocess.call(command, shell=True)
command = "ffmpeg -i " + Input_Video + " -ab 160k -ac 2 -ar " + str(
sample_rate) + " -vn " + TEMP + "/audio.wav"
subprocess.call(command, shell=True)
sampleRate, audioData = wavfile.read(TEMP + "/audio.wav")
audioSampleCount = audioData.shape[0]
maxAudioVolume = GetVolume(audioData)
# print(" please ")
samplesPerFrame = 1470
audioFrameCount = int(math.ceil(audioSampleCount / samplesPerFrame))
hasLoudAudio = numpy.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(i * samplesPerFrame)
end = min(int((i + 1) * samplesPerFrame), audioSampleCount)
audiochunks = audioData[start:end]
maxchunksVolume = float(GetVolume(audiochunks)) / maxAudioVolume
if maxchunksVolume >= silent_threshold:
hasLoudAudio[i] = 1
chunks = [[0, 0, 0]]
shouldIncludeFrame = numpy.zeros((audioFrameCount))
for i in range(audioFrameCount):
start = int(max(0, i - frame_margin))
end = int(min(audioFrameCount, i + 1 + frame_margin))
shouldIncludeFrame[i] = numpy.max(hasLoudAudio[start:end])
if (i >= 1 and shouldIncludeFrame[i] != shouldIncludeFrame[i - 1]):
chunks.append([chunks[-1][1], i, shouldIncludeFrame[i - 1]])
chunks.append(
[chunks[-1][1], audioFrameCount, shouldIncludeFrame[i - 1]])
chunks = chunks[1:]
outputAudioData = numpy.zeros((0, audioData.shape[1]))
outputPointer = 0
lastExistingFrame = None
for chunk in chunks:
audioChunk = audioData[int(chunk[0] *
samplesPerFrame):int(chunk[1] *
samplesPerFrame)]
sFile = TEMP + "/tempStart.wav"
eFile = TEMP + "/tempEnd.wav"
wavfile.write(sFile, sample_rate, audioChunk)
with WavReader(sFile) as reader:
with WavWriter(eFile, reader.channels,
reader.samplerate) as writer:
tsm = phasevocoder(reader.channels,
speed=New_Speed_silent_and_sounded[int(
chunk[2])])
tsm.run(reader, writer)
_, alteredAudioData = wavfile.read(eFile)
leng = alteredAudioData.shape[0]
endPointer = outputPointer + leng
outputAudioData = numpy.concatenate(
(outputAudioData, alteredAudioData / maxAudioVolume))
if leng < Audio_fade_envelope_size:
outputAudioData[outputPointer:endPointer] = 0
else:
premask = numpy.arange(
Audio_fade_envelope_size) / Audio_fade_envelope_size
mask = numpy.repeat(premask[:, numpy.newaxis], 2, axis=1)
outputAudioData[outputPointer:outputPointer +
Audio_fade_envelope_size] *= mask
outputAudioData[
endPointer -
Audio_fade_envelope_size:endPointer] *= 1 - mask
startOutputFrame = int(math.ceil(outputPointer / samplesPerFrame))
endOutputFrame = int(math.ceil(endPointer / samplesPerFrame))
for op_frame in range(startOutputFrame, endOutputFrame):
ip_frame = int(chunk[0] +
New_Speed_silent_and_sounded[int(chunk[2])] *
(op_frame - startOutputFrame))
didItWork = FRameCopy(ip_frame, op_frame)
if didItWork:
lastExistingFrame = ip_frame
else:
FRameCopy(lastExistingFrame, op_frame)
outputPointer = endPointer
wavfile.write(TEMP + "/audioNew.wav", sample_rate, outputAudioData)
command = "ffmpeg -framerate " + str(
frame_rate
) + " -i " + TEMP + "/new_frame%06d.jpg -i " + TEMP + "/audioNew.wav -strict -2 " + Output_Video
subprocess.call(command, shell=True)
try:
rmtree(TEMP, ignore_errors=False)
except OSError:
print("Delete failed")
return "done" #not sure abt it
except:
return " nothing"
def fastVideoPlus(videoFile, outFile, silentThreshold, frameMargin,
SAMPLE_RATE, AUD_BITRATE, VERBOSE, videoSpeed, silentSpeed,
cutByThisTrack, keepTracksSep):
print('Running from fastVideoPlus.py')
if (not os.path.isfile(videoFile)):
print('Could not find file:', videoFile)
sys.exit()
TEMP = '.TEMP'
FADE_SIZE = 400
NEW_SPEED = [silentSpeed, videoSpeed]
cap = cv2.VideoCapture(videoFile)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
fps = round(cap.get(cv2.CAP_PROP_FPS))
try:
os.mkdir(TEMP)
except OSError:
rmtree(TEMP)
os.mkdir(TEMP)
tracks = vidTracks(videoFile)
if (cutByThisTrack >= tracks):
print("Error: You choose a track that doesn't exist.")
print(f'There are only {tracks-1} tracks. (starting from 0)')
sys.exit()
for trackNumber in range(tracks):
cmd = [
'ffmpeg', '-i', videoFile, '-ab', AUD_BITRATE, '-ac', '2', '-ar',
str(SAMPLE_RATE), '-map', f'0:a:{trackNumber}',
f'{TEMP}/{trackNumber}.wav'
]
if (not VERBOSE):
cmd.extend(['-nostats', '-loglevel', '0'])
else:
cmd.extend(['-hide_banner'])
subprocess.call(cmd)
sampleRate, audioData = wavfile.read(f'{TEMP}/{cutByThisTrack}.wav')
chunks = getAudioChunks(audioData, sampleRate, fps, silentThreshold, 2,
frameMargin)
hmm = preview(chunks, NEW_SPEED, fps)
estLeng = int((hmm * SAMPLE_RATE) * 1.5) + int(SAMPLE_RATE * 2)
oldAudios = []
newAudios = []
for i in range(tracks):
__, audioData = wavfile.read(f'{TEMP}/{i}.wav')
oldAudios.append(audioData)
newAudios.append(np.zeros((estLeng, 2), dtype=np.int16))
yPointer = 0
out = cv2.VideoWriter(f'{TEMP}/spedup.mp4', fourcc, fps, (width, height))
channels = 2
switchStart = 0
needChange = False
preve = None
endMargin = 0
yPointer = 0
frameBuffer = []
def writeFrames(frames, nAudio, speed, samplePerSecond, writer):
numAudioChunks = round(nAudio / samplePerSecond * fps)
global nFrames
numWrites = numAudioChunks - nFrames
nFrames += numWrites # if sync issue exists, change this back
limit = len(frames) - 1
for i in range(numWrites):
frameIndex = round(i * speed)
if (frameIndex > limit):
writer.write(frames[-1])
else:
writer.write(frames[frameIndex])
totalFrames = chunks[len(chunks) - 1][1]
outFrame = 0
beginTime = time()
while cap.isOpened():
ret, frame = cap.read()
if (not ret):
break
cframe = int(cap.get(cv2.CAP_PROP_POS_FRAMES)) # current frame
currentTime = cap.get(cv2.CAP_PROP_POS_MSEC) / 1000
audioSampleStart = int(currentTime * sampleRate)
audioSampleEnd = min(
audioSampleStart + sampleRate // fps * frameMargin, len(audioData))
switchEnd = audioSampleStart + sampleRate // fps
audioChunk = audioData[audioSampleStart:audioSampleEnd]
state = None
for chunk in chunks:
if (cframe >= chunk[0] and cframe <= chunk[1]):
state = chunk[2]
break
if (state == 0):
if (endMargin < 1):
isSilent = 1
else:
isSilent = 0
endMargin -= 1
else:
isSilent = 0
endMargin = frameMargin
if (preve is not None and preve != isSilent):
needChange = True
preve = isSilent
if (not needChange):
frameBuffer.append(frame)
else:
theSpeed = NEW_SPEED[isSilent]
if (theSpeed < 99999):
# handle audio tracks
for i, oneAudioData in enumerate(oldAudios):
spedChunk = oneAudioData[switchStart:switchEnd]
spedupAudio = np.zeros((0, 2), dtype=np.int16)
with ArrReader(spedChunk, channels, sampleRate,
2) as reader:
with ArrWriter(spedupAudio, channels, sampleRate,
2) as writer:
phasevocoder(reader.channels,
speed=theSpeed).run(reader, writer)
spedupAudio = writer.output
yPointerEnd = yPointer + spedupAudio.shape[0]
newAudios[i][yPointer:yPointerEnd] = spedupAudio
yPointer = yPointerEnd
else:
yPointerEnd = yPointer
writeFrames(frameBuffer, yPointerEnd, NEW_SPEED[isSilent],
sampleRate, out)
frameBuffer = []
switchStart = switchEnd
needChange = False
progressBar(cframe, totalFrames, beginTime)
# finish audio
for i, newData in enumerate(newAudios):
newData = newData[:yPointer]
wavfile.write(f'{TEMP}/new{i}.wav', sampleRate, newData)
if (not os.path.isfile(f'{TEMP}/new{i}.wav')):
raise IOError('audio file not created.')
cap.release()
out.release()
cv2.destroyAllWindows()
first = videoFile[:videoFile.rfind('.')]
extension = videoFile[videoFile.rfind('.'):]
if (outFile == ''):
outFile = f'{first}_ALTERED{extension}'
# Now mix new audio(s) and the new video.
if (keepTracksSep):
cmd = ['ffmpeg', '-y']
for i in range(tracks):
cmd.extend(['-i', f'{TEMP}/new{i}.wav'])
cmd.extend(['-i', f'{TEMP}/spedup.mp4']) # add input video
for i in range(tracks):
cmd.extend(['-map', f'{i}:a:0'])
cmd.extend([
'-map', f'{tracks}:v:0', '-c:v', 'copy', '-movflags', '+faststart',
outFile
])
if (not VERBOSE):
cmd.extend(['-nostats', '-loglevel', '0'])
else:
if (tracks > 1):
cmd = ['ffmpeg']
for i in range(tracks):
cmd.extend(['-i', f'{TEMP}/new{i}.wav'])
cmd.extend([
'-filter_complex', f'amerge=inputs={tracks}', '-ac', '2',
f'{TEMP}/newAudioFile.wav'
])
if (not VERBOSE):
cmd.extend(['-nostats', '-loglevel', '0'])
subprocess.call(cmd)
else:
os.rename(f'{TEMP}/new0.wav', f'{TEMP}/newAudioFile.wav')
cmd = [
'ffmpeg', '-y', '-i', f'{TEMP}/newAudioFile.wav', '-i',
f'{TEMP}/spedup.mp4', '-c:v', 'copy', '-movflags', '+faststart',
outFile
]
if (not VERBOSE):
cmd.extend(['-nostats', '-loglevel', '0'])
subprocess.call(cmd)
return outFile
# coding: utf-8
# In[11]:
from audiotsm import phasevocoder
from audiotsm.io.wav import WavReader, WavWriter
with WavReader('qbhexamples.wav') as reader:
print reader.channels, reader.samplerate
with WavWriter('qbh_half.wav', reader.channels, reader.samplerate) as writer:
tsm = phasevocoder(reader.channels, speed=0.5)
tsm.run(reader, writer)
print "Finished, closing files."
close(reader)
close(writer)
# In[ ]: