def _get_track_metadata(path):
# use mono at 44kHz as reference. For any other settings data won't be perfectly
# normalized but it should be good enough.
audio = AudioFile(path)
mix = audio.read(streams=0, channels=1, samplerate=44100)
return {
"duration": audio.duration,
"std": mix.std().item(),
"mean": mix.mean().item()
}
def test_recognize(key, filename):
"""
Microsoft Bing Speech APIを使って音声認識するテスト用関数
"""
from audio import AudioData, AudioFile
af = AudioFile(filename)
af.__enter__()
ad = AudioData(
open(filename, 'rb').read(), af.SAMPLE_RATE, af.SAMPLE_WIDTH)
bs = Bing()
r = bs.recognize(ad, key=key, show_all=True)
print(r)
def test_recognize(key, filename):
"""
Google Cloud Speech APIを使って音声認識するテスト用関数
"""
from audio import AudioData, AudioFile
af = AudioFile(filename)
af.__enter__()
ad = AudioData(
open(filename, 'rb').read(), af.SAMPLE_RATE, af.SAMPLE_WIDTH)
gs = Google()
r = gs.recognize(ad, key=key, show_all=True)
print(r)
def __getitem__(self, index):
track = self.mus.tracks[index]
return (track.name,
AudioFile(track.path).read(channels=self.channels,
seek_time=0,
streams=self.streams,
samplerate=self.samplerate))
def censor(self):
""" Creates a clean/new version of a file by removing explicits """
audio_file = AudioFile(self.file_path)
# Define the CLI progress bar
p_bar, p_bar_step = self.__progress_bar(audio_file.normal_chunks)
async_iter = zip(repeat(p_bar), repeat(p_bar_step),
audio_file.normal_chunks)
# Censor each audio chunk file asynchronously
censored_chunks = ThreadPool(6).map(self.__censor_chunk, async_iter)
clean_file = self.__create_clean_segment(censored_chunks)
p_bar.close()
self.create_clean_file(clean_file)
def __getitem__(self, index):
for meta in self.metadata:
examples = self._examples_count(meta)
if index >= examples:
index -= examples
continue
streams = AudioFile(meta["path"]).read(seek_time=index *
self.stride,
duration=self.duration,
channels=self.channels,
samplerate=self.samplerate)
return (streams - meta["mean"]) / meta["std"]
def transcribe_file(audio_path):
# split_audio_file parameter variable settings
batchsize = 128
# Aggressiveness has to be within 0-3 for VAD split to work.
aggressive = 1
outlierdurms = 20
outlierbatchsize = 64
# from .... import DS model
# Makes sure there are cores present in a system to be used
try:
num_processes = cpu_count()
except NotImplementedError:
num_processes = 1
with AudioFile(audio_path, as_path=True) as wav_path:
dataset = split_audio_file(wav_path,
batch_size=batchsize,
aggressiveness=aggressive,
outlier_duration_ms=outlierdurms,
outlier_batch_size=outlierbatchsize)
print("dataset:", dataset)