def determineSilences(file):
trackA = AudioSegment.from_file(file["trackA"])
speakingA = silence.detect_nonsilent(trackA, min_silence_len=500, silence_thresh=-30)
trackB = AudioSegment.from_file(file["trackB"])
speakingB = silence.detect_nonsilent(trackB, min_silence_len=500, silence_thresh=-30)
aClass = [[set[0],set[1], hasConflicts(set, speakingB)] for set in speakingA]
bClass = [[set[0], set[1], hasConflicts(set, speakingA)] for set in speakingB]
return aClass, bClass
def compareSpeechPeriods(self, source_in, source_out):
segment_in = AudioSegment(data=source_in,
sample_width=2,
frame_rate=8000,
channels=1)
segment_out = AudioSegment(data=source_out,
sample_width=2,
frame_rate=8000,
channels=1)
speech_period_in = silence.detect_nonsilent(segment_in,
min_silence_len=1000,
silence_thresh=-32)
speech_period_out = silence.detect_nonsilent(segment_out,
min_silence_len=1000,
silence_thresh=-32)
#print('period in: ', speech_period_in)
#print('period out: ', speech_period_out)
if len(speech_period_out) == 0 and len(speech_period_in) == 0:
return 0
elif len(speech_period_out) == 0 and len(speech_period_in) > 0:
voiceEdgePoints = [
speech_period_in[0][0],
speech_period_in[len(speech_period_in) - 1][1]
]
return (1, voiceEdgePoints)
elif len(speech_period_in) == 0 and len(speech_period_out) > 0:
voiceEdgePoints = [
speech_period_out[0][0],
speech_period_out[len(speech_period_out) - 1][1]
]
return (2, voiceEdgePoints)
elif speech_period_in[0][0] <= speech_period_out[0][0]:
if speech_period_out[len(speech_period_out) -
1][1] >= speech_period_in[
len(speech_period_in) - 1][1]:
voiceEdgePoint = speech_period_out[len(speech_period_out) -
1][1]
else:
voiceEdgePoint = speech_period_in[len(speech_period_in) - 1][1]
voiceEdgePoints = [speech_period_in[0][0], voiceEdgePoint]
return (3, voiceEdgePoints)
elif speech_period_in[0][0] > speech_period_out[0][0]:
if speech_period_out[len(speech_period_out) -
1][1] >= speech_period_in[
len(speech_period_in) - 1][1]:
voiceEdgePoint = speech_period_out[len(speech_period_out) -
1][1]
else:
voiceEdgePoint = speech_period_in[len(speech_period_in) - 1][1]
voiceEdgePoints = [speech_period_out[0][0], voiceEdgePoint]
return (4, voiceEdgePoints)
def trim_silence(wav):
start_end = detect_nonsilent(wav, 250, -40, 1)
start_end = [se for se in start_end if se[1] - se[0] > 50]
while len(start_end) == 0 or wav.dBFS > 0: #if can't detect nonsilent
wav = wav + 5
start_end = detect_nonsilent(wav, 250, -40, 1)
start_end = [se for se in start_end if se[1] - se[0] > 50]
start = min(start_end)[0]
end = max(start_end)[1]
wav = wav[start - 50:end + 50]
wav = np.array(wav.get_array_of_samples(), dtype='float32') / 2**15
return wav
def goAmadeus(file,targetFolder, silenceModifier):
audio_segment = AudioSegment.from_wav(file)
normalized_sound = match_target_amplitude(audio_segment, -20.0)
nonsilent_data = detect_nonsilent(normalized_sound, min_silence_len=50, silence_thresh=-45, seek_step=1)
# print("start,Stop")
# for chunks in nonsilent_data:
# print([chunk / 1000 for chunk in chunks])
activeHolder = 0
where = 1
howManyLoops = 0
for i in range(len(nonsilent_data)):
if(where-1 < 0):
activeHolder += 0
where += 0
if where >= 1:
current_silence = nonsilent_data[howManyLoops][0] - nonsilent_data[howManyLoops-1][1]
if(current_silence >= 2000):
activeHolder += 0
elif current_silence <= 50 and howManyLoops < 3:
activeHolder += 5
elif current_silence > 50:
activeHolder += current_silence
where += 1
if(where == 0):
where += 1
howManyLoops += 1
allSilence = activeHolder
# print("All silence: " + str(allSilence) + " ms")
bestSingleSilence = allSilence / where
add = bestSingleSilence * silenceModifier/100
bestSingleSilence = bestSingleSilence+add
# print("Best Silence: " + str(round(bestSingleSilence)) + " ms")
# print("Started Chunking..")
cutOnBestSilence(round(bestSingleSilence), file, targetFolder)
best_nonsilent_data = detect_nonsilent(normalized_sound, min_silence_len=round(bestSingleSilence), silence_thresh=-45, seek_step=1)
return best_nonsilent_data
# goAmadeus("test.wav",1)
def run(self):
try:
if os.path.exists(self.wav_name) is not True:
AudioSegment.from_file(self.video_name).export(self.wav_name,
format='mp3',
bitrate="64k")
sound = AudioSegment.from_mp3(self.wav_name)
chunks = detect_nonsilent(sound,
min_silence_len=1000,
silence_thresh=-45)
# now recombine the chunks so that the parts are at least 60 sec long
target_length = 60 * 1000
output_chunks = [chunks[0]]
for chunk in chunks[1:]:
if output_chunks[-1][1] - output_chunks[-1][0] < target_length:
output_chunks[-1][1] = chunk[1]
else:
# if the last output chunk is longer than the target length,
# we can start a new one
output_chunks.append(chunk)
config = {
'duration': sound.duration_seconds,
'total': len(output_chunks),
'chunks': output_chunks
}
self.config_sig.emit(config)
except Exception as e:
print(e)
self.config_sig.emit({})
def get_first_nonsilent(self, sound, silence_threshold=-28):
non_silences = silence.detect_nonsilent(
sound,
min_silence_len=self.fade_length,
silence_thresh=silence_threshold,
seek_step=1)
return non_silences[0]
def cut_speech(dirs):
"""
This function is used to cut speech samples into non-silent pieces for further usage in MFCC extraction
:return: None
"""
logging.basicConfig(filename="logfile.log", level=logging.DEBUG)
for d in dirs:
for speaker in os.listdir(d):
wav_names = glob('{}/{}/*.wav'.format(d, speaker))
for name in wav_names:
print "processing file {}...".format(name)
sound = AudioSegment.from_file(name, format="wav")
speech = silence.detect_nonsilent(sound, silence_thresh=-50)
i = 1
for frag in speech:
part = sound[frag[0]:frag[1]]
part.export('{}/{}/part_{}.wav'.format(d, speaker, i),
format="wav")
i += 1
logging.info("Finished cutting audio samples")
def splitAudioBySilence(audio_path,
skip_idx=0,
out_ext="wav",
silence_thresh=-40,
silence_chunk_len=100,
keep_silence=100):
audio = read_audio(audio_path)
not_silence_ranges = silence.detect_nonsilent(
audio,
min_silence_len=silence_chunk_len,
silence_thresh=silence_thresh)
edges = concatenate_edges(not_silence_ranges)
intervals = get_rid_of_short_intervals(edges)
for idx, (start_idx, end_idx) in enumerate(intervals[skip_idx:]):
start_idx = max(0, start_idx - keep_silence)
end_idx += keep_silence
segment = audio[start_idx:end_idx]
segment.export("./chunks/chunk{0}.mp3".format(idx), out_ext)
segment.set_channels(1)
segment.export("./chunks/chunk{0}.wav".format(idx), format="wav")
os.remove("./chunks/chunk{0}.mp3".format(idx))
# emptyFolder("./chunks/")
# splitAudioBySilence("audio.mp3")
def split_audio(self,
audio_segment,
min_silence_len=500,
silence_thresh=-30,
keep_silence=100,
seek_step=1):
not_silence_ranges = detect_nonsilent(audio_segment, min_silence_len,
silence_thresh, seek_step)
def pairwise(iterable):
"s -> (s0,s1), (s1,s2), (s2, s3), ..."
a, b = itertools.tee(iterable)
next(b, None)
return zip(a, b)
chunks = []
audio_starts = []
audio_ends = []
# 根据silence片段切分视频
for (start1, end1), (start2, end2) in pairwise(not_silence_ranges):
chunks.append(audio_segment[end1:start2])
audio_starts.append(end1)
audio_ends.append(start2)
return chunks, audio_starts, audio_ends
def get_segments(args, audio_file, segments_file):
if os.path.exists(segments_file):
with open(segments_file) as json_file:
json_data = json.load(json_file)
return json_data["sound_ranges"][0]["sounds"]
else:
long_nonsilence = detect_nonsilent(audio_file,
min_silence_len=args.long_silence,
silence_thresh=args.silence_thresh)
silence = detect_silence(audio_file,
min_silence_len=args.short_silence,
silence_thresh=args.silence_thresh)
gaps_silence = list(
map(
lambda x:
[x[0] + args.short_silence / 2, x[1] - args.short_silence / 2],
detect_silence(audio_file,
min_silence_len=2 * args.short_silence,
silence_thresh=args.silence_thresh + 20)))
nonsilence1 = split_long_nonsilence(long_nonsilence, silence,
args.min * 1000, args.max * 1000)
segments = split_long_nonsilence(nonsilence1, gaps_silence,
args.min * 1000, args.max * 1000)
return segments
def get_non_silent_ranges(filepath, audio_length, silence_length,
silence_thresh):
"""
Given a filepath to a .wav file and a target audio length, return all the
non-silent ranges from the audio sample
:param filepath: filepath to the .wav audio file
:param audio_length: length in seconds of audio to process
:param silence_length: minimum length of a silence to be used for
a split
:param silence_thresh: (in dBFS) anything quieter than this will be
considered silence
:return: 2D array: array of shape [num_ranges, 2]
"""
# Load data into AudioSegment object and extract audio_length seconds
audio_file = AudioSegment.from_wav(filepath)
audio_file = audio_file[:audio_length * MS]
# Use given parameters to return non_silent ranges
ranges = []
for i in range(10):
ranges = detect_nonsilent(audio_file,
min_silence_len=int(silence_length),
silence_thresh=silence_thresh)
if len(ranges) > 0:
break
else:
silence_length /= 2
return ranges
def cut_by_silence(precut_audio_path, output_folder, filebasename):
if not os.path.exists(output_folder):
os.mkdir(output_folder)
#use pydub AudioSegment &silence module to detect silence, cut and save,
#return last chunck's time stamp in millionsecond
cut_num = 0
audio_segment = AudioSegment.from_wav(precut_audio_path)
silence_thresh_tries = range(-40, -5)
for silence_thresh in silence_thresh_tries:
chuncks = detect_nonsilent(audio_segment,
min_silence_len=500,
silence_thresh=silence_thresh)
logging.debug("try {}".format(silence_thresh))
if len(chuncks) >= 2:
for chunck in chuncks:
out_audio_file = os.path.join(
output_folder, filebasename + "_" +
str(TimestampMillisec64()) + "_" + str(cut_num) + ".wav")
audio_segment[chunck[0]:chunck[1]].export(out_audio_file,
format='wav')
cut_num = cut_num + 1
break
if silence_thresh == -5 and len(chuncks) < 2:
out_audio_file = os.path.join(
output_folder, filebasename + "_" + str(TimestampMillisec64()) +
"_" + str(cut_num) + ".wav")
audio_segment[chuncks[0][0]:chuncks[0][1]].export(out_audio_file,
format='wav')
return 60, cut_num
def preprocess_wav_files(my_dir):
vowelsDict = defaultdict(list)
for filename in os.listdir(my_dir):
vowel = filename.split("_")[0][-1]
# File read
signal = AudioSegment.from_wav(my_dir + '/' + filename)
# Remove silence - beginning and end
non_sil_times = detect_nonsilent(signal,
min_silence_len=50,
silence_thresh=signal.dBFS * 1.5)
if len(non_sil_times):
signal = signal[non_sil_times[0][0]:non_sil_times[0][1]]
# Downsampling to 16KHz
signal = signal.set_frame_rate(RATE)
# Wav segmentation
segmented_signal = wav_segmentation(signal)
segmented_signal = [
chunk.get_array_of_samples() for chunk in segmented_signal
]
if '_' in filename:
vowelsDict[vowel].extend(segmented_signal)
# librosa_data[vowel].append(librosa_features(my_dir + '/' + filename))
else:
vowelsDict['iau'].extend(segmented_signal)
return vowelsDict
def splitSound(self):
self.__soundChunks = []
gaps = silence.detect_nonsilent(self.__audio,
silence_thresh=self.__threshold)
for start, final in gaps:
self.__soundChunks.append(self.__audio[start:final])
return self
def trim_on_silence(audio_path,
skip_idx=0,
out_ext="wav",
silence_thresh=-40,
min_silence_len=400,
silence_chunk_len=100,
keep_silence=200):
audio = read_audio(audio_path)
not_silence_ranges = silence.detect_nonsilent(
audio,
min_silence_len=silence_chunk_len,
silence_thresh=silence_thresh)
if not not_silence_ranges:
print(audio_path)
return []
start_idx = not_silence_ranges[0][0]
end_idx = not_silence_ranges[-1][1]
start_idx = max(0, start_idx - keep_silence)
end_idx = min(len(audio), end_idx + keep_silence)
trimmed = audio[start_idx:end_idx]
trimmed.export(audio_path, out_ext)
return []
def split_on_silence_with_pydub(
audio_path, skip_idx=0, out_ext="wav",
silence_thresh=-40, silence_chunk_len=100, keep_silence=100):
filename = os.path.basename(audio_path).split('.', 1)[0]
audio = read_audio(audio_path)
not_silence_ranges = silence.detect_nonsilent(
audio, min_silence_len=silence_chunk_len,
silence_thresh=silence_thresh)
edges = concatenate_edges(not_silence_ranges)
intervals = get_rid_of_short_intervals(edges)
# Save audio files
audio_paths = []
for idx, (start_idx, end_idx) in enumerate(intervals[skip_idx:]):
start_idx = max(0, start_idx - keep_silence)
end_idx += keep_silence
target_audio_path = "{}/pre_audio/{}.{:04d}.{}".format(os.path.dirname(audio_path), filename, idx, out_ext)
segment = audio[start_idx:end_idx]
segment.export(target_audio_path, out_ext)
audio_paths.append(target_audio_path)
return audio_paths, intervals
def extractor(self):
seg = AudioSegment.from_file(self.song)
# reduce loudness of sounds over 120Hz (focus on bass drum, etc)
seg = seg.low_pass_filter(120.0)
# we'll call a beat: anything above average loudness
beat_loudness = seg.dBFS
# the fastest tempo we'll allow is 240 bpm (60000ms / 240beats)
minimum_silence = int(60000 / 240.0)
nonsilent_times = detect_nonsilent(seg, minimum_silence, beat_loudness)
spaces_between_beats = []
last_t = nonsilent_times[0][0]
for peak_start, _ in nonsilent_times[1:]:
spaces_between_beats.append(peak_start - last_t)
last_t = peak_start
# We'll base our guess on the median space between beats
spaces_between_beats = sorted(spaces_between_beats)
space = spaces_between_beats[int(len(spaces_between_beats) / 2)]
bpm = 60000 / space
return bpm
def _split_on_silence_ranges(self,
min_silence_len=1000,
silence_thresh=-16,
keep_silence=100,
seek_step=1):
# from the itertools documentation
def pairwise(iterable):
"s -> (s0,s1), (s1,s2), (s2, s3), ..."
a, b = itertools.tee(iterable)
next(b, None)
return zip(a, b)
if isinstance(keep_silence, bool):
keep_silence = len(self.audio) if keep_silence else 0
output_ranges = [[
start - keep_silence, end + keep_silence
] for (start, end) in detect_nonsilent(self.audio, min_silence_len,
silence_thresh, seek_step)]
for range_i, range_ii in pairwise(output_ranges):
last_end = range_i[1]
next_start = range_ii[0]
if next_start < last_end:
range_i[1] = (last_end + next_start) // 2
range_ii[0] = range_i[1]
return [(max(start, 0), min(end, len(self.audio)))
for start, end in output_ranges]
def split_videos(filename, output):
global silence
# Split file on silences
video = AudioSegment.from_file(filename, "mp4")
silences = silence.detect_nonsilent(video,
silence_thresh=-40,
min_silence_len=400)
print("Done splitting audio, start splitting clips")
print(filename)
clip = VideoFileClip(filename)
for i, silence in enumerate(silences):
start = silence[0] / 1000
end = silence[1] / 1000
length = end - start
long_vid = False
while length > 10:
long_vid = True
temp = start + 5
subpart = clip.subclip(start, temp)
subpart.write_videofile("{}/000s{}.mp4".format(output, i))
i += 1
start += 5
length -= 5
if long_vid or length < 1:
continue
subpart = clip.subclip(start, end)
subpart.write_videofile("{}/000{}.mp4".format(output, i))
def splitAudioFile(filename_mp3, in_dir, min_silence_len=400, silence_thresh=-65):
sound = AudioSegment.from_mp3(filename_mp3)
nonsilence_range = detect_nonsilent(sound, min_silence_len, silence_thresh)
chunks_range = []
for i, chunk in enumerate(nonsilence_range):
if i==0:
print(chunk, len(nonsilence_range))
start = chunk[0]
end = (chunk[1] + nonsilence_range[i+1][0])/2
sound[:end].export(".\\tmp\\%d\\%d.wav" % (in_dir, i), format="wav", bitrate="16k")
elif i == len(nonsilence_range)-1:
start = (nonsilence_range[i-1][1] + chunk[0])/2
end = chunk[1] + 1000.0
sound[start:].export(".\\tmp\\%d\\%d.wav" % (in_dir, i), format="wav", bitrate="16k")
else:
start = (nonsilence_range[i-1][1] + chunk[0])/2
end = (chunk[1] + nonsilence_range[i+1][0])/2
sound[start:end].export(".\\tmp\\%d\\%d.wav" % (in_dir, i), format="wav", bitrate="16k")
start = round(start/1000, 1)
end = round(end/1000, 1)
chunks_range.append((start,end))
return chunks_range
def split_on_silence(audio_segment,
min_silence_len=1000,
silence_thresh=-16,
keep_silence=100,
seek_step=1):
"""
audio_segment - original pydub.AudioSegment() object
min_silence_len - (in ms) minimum length of a silence to be used for
a split. default: 1000ms
silence_thresh - (in dBFS) anything quieter than this will be
considered silence. default: -16dBFS
keep_silence - (in ms) amount of silence to leave at the beginning
and end of the chunks. Keeps the sound from sounding like it is
abruptly cut off. (default: 100ms)
"""
not_silence_ranges = detect_nonsilent(audio_segment, min_silence_len,
silence_thresh, seek_step)
chunks = []
starttime = []
endtime = []
for start_i, end_i in not_silence_ranges:
start_i = max(0, start_i - keep_silence)
end_i += keep_silence
chunks.append(audio_segment[start_i:end_i])
starttime.append(start_i)
endtime.append(end_i)
return chunks, starttime, endtime
def get_segmented_samples(my_dir, dictsList, iter_num=5):
all_pitchs = []
for dictL in dictsList:
all_speakers = []
for speakerNum in dictL.keys():
segments = []
for file in dictL[speakerNum]:
signal = AudioSegment.from_wav(my_dir + '\\' + file)
# Remove silence - beginning and end
non_sil_times = detect_nonsilent(signal, min_silence_len=50, silence_thresh=signal.dBFS * 1.5)
if len(non_sil_times): signal = signal[non_sil_times[0][0]:non_sil_times[0][1]]
# Downsampling to 16KHz
signal = signal.set_frame_rate(RATE)
# Wav segmentation
segmented_signal = wav_segmentation(signal)
segmented_signal = [chunk.get_array_of_samples() for chunk in segmented_signal]
segments.append(segmented_signal)
iter_segments_per_speaker = []
for n in range(iter_num):
tmp = []
for segList in segments:
rand_idx = np.random.randint(len(segList))
tmp.append(segList[rand_idx])
iter_segments_per_speaker.append(tmp)
all_speakers.append(iter_segments_per_speaker)
all_pitchs.append(all_speakers)
return all_pitchs
def shorter_filler(json_result, audio_file, min_silence_len, start_time, non_silence_start):
# 침묵 길이를 더 짧게
min_silence_length = (int)(min_silence_len/1.2)
intervals = detect_nonsilent(audio_file,
min_silence_len=min_silence_length,
silence_thresh=-32.64
)
for interval in intervals:
interval_audio = audio_file[interval[0]:interval[1]]
# padding 40 길이 이상인 경우 더 짧게
if (interval[1]-interval[0] >= 460):
non_silence_start = shorter_filler(json_result, interval_audio, min_silence_length, interval[0]+start_time, non_silence_start)
else:# padding 40 길이보다 짧은 경우 predict
if interval[1]-interval[0] > 10 :
if predict_filler(interval_audio) == 0 : # 추임새인 경우
json_result.append({'start':non_silence_start,'end':start_time+interval[0],'tag':'1000'}) # tag: 1000 means non-slience
non_silence_start = start_time + interval[0]
# 추임새 tagging
json_result.append({'start':start_time+interval[0],'end':start_time+interval[1],'tag':'1111'}) # tag: 1111 means filler word
return non_silence_start
def volume_trial(file):
# The threshhold of volume (decibels)
threshhold = -45
# The threshhold of total time over volume threshhold (seconds)
length_seconds = 5
# Get audio from file and data about audio
audio = AudioSegment.from_file(file)
data = mediainfo(file)
# The number of samples in the audio per second of audio
samples_per_second = int(len(audio) / float(data["duration"]))
# The time threshhold in samples
length_samples = length_seconds * samples_per_second
# Find all sets of nonsilent samples
nonsilences = silence.detect_nonsilent(audio, 1, silence_thresh=threshhold)
# calculate total amount of nonsilences
total = 0
for i in nonsilences:
total += i[1] - i[0]
print(total / samples_per_second)
if total >= length_samples:
return True
else:
return False
def calc_bpm(audio: AudioSegment) -> int:
tmp_audio = effects.low_pass_filter(audio,
120) # cut off sounds below 120 Hz
beat_volume = tmp_audio.dBFS
min_silence = int(60000 / 240.0) # Allow up to 240 bpm
nonsilent_ranges = detect_nonsilent(tmp_audio, min_silence, beat_volume)
spaces_between_beats = []
last_t = nonsilent_ranges[0][0]
for peak_start, _ in nonsilent_ranges[1:]:
spaces_between_beats.append(peak_start - last_t)
last_t = peak_start
spaces_between_beats = sorted(spaces_between_beats)
temp = len(spaces_between_beats) / 2
temp = int(temp)
print(temp)
if temp == 0:
# This just means that this segment had no audio louder then 120 Hz
# might as well discard it since we aren't going to get a good bpm
# measurment from quieter sections of the song
return 0
space = spaces_between_beats[temp]
bpm = 60000 / space
return bpm
def trim_silent(self):
""" Trims extraneous silence at the ends of the audio """
a = AudioSegment.empty()
for seg in silence.detect_nonsilent(self.audio):
a = a.append(self.audio[seg[0]:seg[1]], crossfade=0)
self.audio = a
return self
def chunk(self):
print('In Audio_Object.Chunk')
''' Chunk the audio in smaller files and save to chunk_path '''
# open the audio file using pydub
sound = AudioSegment.from_wav(self.audio_path +
self.audio_filename +
self.audio_extn)
silence_to_keep = 100
# split audio sound where silence is 700 miliseconds or more and get chunks
chunks = split_on_silence(sound,
# experiment with this value for your target audio file
min_silence_len = 700,
# adjust this per requirement
silence_thresh = sound.dBFS-14,
# keep the silence for 1 second, adjustable as well
keep_silence=silence_to_keep,
)
# create a directory to store the audio chunks
if not os.path.isdir(self.chunk_path):
os.mkdir(self.chunk_path)
times = []
#Print detected non-silent chunks, which in our case would be spoken words.
nonsilent_data = detect_nonsilent(sound, min_silence_len=700, silence_thresh=sound.dBFS-14, seek_step=1)
#convert ms to seconds
print("start,Stop")
for chunks_times in nonsilent_data:
times.append( [ct/1000 for ct in chunks_times])
self.ts = pd.DataFrame(times, columns=["start_time", "stop_time"])
# print(self.ts.head())
# process each chunk
dur_sec = []
text = []
for i, audio_chunk in enumerate(chunks, start=1):
# export audio chunk and save it in
# the `folder_name` directory.
chunk_filename = os.path.join(self.chunk_path, f"chunk{i}.wav")
audio_chunk.export(chunk_filename, format="wav")
dur_sec.append(audio_chunk.duration_seconds -
((silence_to_keep*2)/1000))
text.append(Audio_Object.extract_text(chunk_filename))
# return the text for all chunks detected
# return(self.whole_text)
self.ts['duration'] = dur_sec
self.ts['text'] = text
self.ts['label'] = self.audio_filename
self.save_ts_df()
def detect_nonsilences(self, sound):
snd = AudioSegment.from_wav(sound)
dBFS = snd.dBFS
non_silent = detect_nonsilent(snd,
min_silence_len=1000,
silence_thresh=dBFS - 16)
#Convert to seconds
non_silence = [((start / 1000), (stop / 1000))
for start, stop in non_silent]
return non_silence
def cutspeech(self, song1):
not_silence_ranges = detect_nonsilent(song1,
min_silence_len=100,
silence_thresh=-32)
starti = not_silence_ranges[0][0]
if len(not_silence_ranges) == 0:
return song1
endi = not_silence_ranges[-1][1]
return song1[starti:endi]
def detect_nonsilence_audiotime(filename, format):
sound = AudioSegment.from_file(filename, format=format)
dbfs = sound.dBFS
print("正在查找非静音片段，根据视频大小，等待时间会有不同～请耐心等待")
timestamp_list = detect_nonsilent(sound,
min_silence_len=700,
silence_thresh=dbfs - 16,
seek_step=1)
print('恭喜你，已经完成' + "共找到" + str(len(timestamp_list)) + "段语音")
return timestamp_list
def split_on_silence_with_pydub(
audio_path, deepspeech, skip_idx=0, out_ext="wav",
silence_thresh=-40, min_silence_len=400,
silence_chunk_len=100, keep_silence=100,
min_segment_length=0): #silence_chunk_len 100->200
filename = os.path.basename(audio_path).split('.', 1)[0]
in_ext = audio_path.rsplit(".")[1]
audio = read_audio(audio_path)
audio= audio.set_channels(1)
audio_sample_width=audio.sample_width
min_chunk_len=int(float(1000)*min_segment_length)
not_silence_ranges = silence.detect_nonsilent(
audio, min_silence_len=silence_chunk_len,
silence_thresh=silence_thresh)
edges = [not_silence_ranges[0]]
for idx in range(1, len(not_silence_ranges)-1):
cur_start = not_silence_ranges[idx][0]
prev_end = edges[-1][1]
prev_start = edges[-1][0]
prev_len = prev_end - prev_start
# if silence is too short or nonsilent is too short
# merge current nonsilence with prev one
if cur_start - prev_end < min_silence_len or (min_chunk_len!=0 and prev_len < min_chunk_len) :
edges[-1][1] = not_silence_ranges[idx][1]
else:
edges.append(not_silence_ranges[idx])
print("Finished finding Edges")
audio_paths = []
for idx, (start_idx, end_idx) in enumerate(edges[skip_idx:]):
start_idx = max(0, start_idx - keep_silence)
end_idx += keep_silence
target_audio_path = "{}/{}.{:04d}.{}".format(
os.path.dirname(audio_path), filename, idx, out_ext)
segment=audio[start_idx:end_idx]
# Set this to deepspeech compatible
if deepspeech:
temp=segment.set_frame_rate(16000)
temp=temp.set_sample_width(2)
temp=temp.set_channels(1)
temp.export(target_audio_path, out_ext) # for soundsegment
else:
segment.export(target_audio_path, out_ext) # for soundsegment
audio_paths.append(target_audio_path)
return audio_paths
def bpm(seg):
l_seg = seg.low_pass_filter(120.0)
beat_loudness = l_seg.dBFS
minimum_silence = int(60000/240.0)
nonsilent_times = detect_nonsilent(l_seg, minimum_silence, beat_loudness)
spaces_between_beats = []
last_t = nonsilent_times[0][0]
for peak_start, _ in nonsilent_times[1:]:
spaces_between_beats.append(peak_start - last_t)
last_t = peak_start
spaces_between_beats = sorted(spaces_between_beats)
space = spaces_between_beats[len(spaces_between_beats) / 2]
bpm = 60000 / space
return bpm
def process(self, dl_entry):
"""Make new audio file in the media directory.
Take the audio file pointed to by dl_entry, normalize, remove silence,
convert to output_format.
"""
input_format = dl_entry.file_extension.lstrip('.')
try:
loader = load_functions[input_format]
except KeyError:
loader = lambda file: AudioSegment.from_file(
file=file, format=input_format)
segment = loader(dl_entry.file_path) # This
# sometimes raised a pydub.exceptions.CouldntDecodeError
segment = segment.normalize() # First normalize
# Try to remove silence
loud_pos = detect_nonsilent(
segment, min_silence_len=minimum_silence_length,
silence_thresh=silence_threshold)
fade_in_length = rapid_fade_length
fade_out_length = rapid_fade_length
if len(loud_pos) == 1:
loud_p = loud_pos[0]
if loud_p[0] > silence_fade_length:
fade_in_length = silence_fade_length
if loud_p[1] < len(segment) - silence_fade_length:
fade_out_length = silence_fade_length
if loud_p[0] > 0 or loud_p[1] < len(segment):
segment = segment[loud_p[0] : loud_p[1]]
segment = segment.fade_in(fade_in_length).fade_out(fade_out_length)
# Now write
tof = tempfile.NamedTemporaryFile(
delete=False, suffix=output_suffix, prefix=u'anki_audio_')
temp_out_file_name = tof.name
tof.close()
segment.export(temp_out_file_name, output_format)
os.unlink(dl_entry.file_path) # Get rid of unprocessed version
return temp_out_file_name, output_suffix
