def convert_audio_miniaudio(in_file, out_file):
channels = 1
sample_rate = 44100
src = miniaudio.decode_file(
in_file) # , dither=miniaudio.DitherMode.TRIANGLE
# DecodedSoundFile - Contains various properties and also the PCM frames of
# a fully decoded audio file.
tgt = miniaudio.DecodedSoundFile("result", 1, sample_rate,
miniaudio.SampleFormat.SIGNED16,
array.array('b'))
converted_frames = miniaudio.convert_frames(src.sample_format,
src.nchannels, src.sample_rate,
src.samples.tobytes(),
tgt.sample_format,
tgt.nchannels, tgt.sample_rate)
tgt.num_frames = int(
len(converted_frames) / tgt.nchannels / tgt.sample_width)
tgt.samples.frombytes(converted_frames)
miniaudio.wav_write_file(out_file, tgt)
logger.debug("wrote converted file to '%s'" % out_file)
return False
def transcribe_file():
if request.method == 'POST':
# Authenticate first
print("INFO] token", request.form.get("token"))
authResponse = requests.post('http://127.0.0.1:8001/next',
json={
"token": request.form.get("token")
}).json()
if "token" not in authResponse:
return jsonify("Forbidden")
# upload wav_file to work directory
f = request.files['file']
greedy = True
if request.form.get('beam'):
if not ENABLE_NGRAM:
print(
"Error: Beam Search with ngram LM is not enabled on this server"
)
else:
greedy = False
file_path = os.path.join(
WORK_DIR, secure_filename(str(uuid.uuid1()) + f.filename))
f.save(file_path)
# conversion to support
inputAudio = miniaudio.wav_read_file_s16(file_path)
result = miniaudio.convert_frames(inputAudio.sample_format, inputAudio.nchannels, \
inputAudio.sample_rate, bytes(inputAudio.samples), \
miniaudio.SampleFormat.UNSIGNED8, 1, 16000)
newFileName = secure_filename(str(uuid.uuid1()) + 'converted.wav')
newFilePath = os.path.join(WORK_DIR, newFileName)
outputAudio = miniaudio.DecodedSoundFile(
newFileName, 1, 16000, miniaudio.SampleFormat.UNSIGNED8,
array.array('B', result))
miniaudio.wav_write_file(newFilePath, outputAudio)
# create manifest
manifest = dict()
manifest['audio_filepath'] = newFilePath
manifest['duration'] = 18000
manifest['text'] = 'todo'
with open(newFilePath + ".json", 'w') as fout:
fout.write(json.dumps(manifest))
start_t = time.time()
transcription = wav_to_text(newFilePath + ".json", greedy=greedy)
total_t = time.time() - start_t
os.remove(file_path)
os.remove(newFileName)
os.remove(newFilePath + ".json")
return jsonify({
"transcription": transcription,
"total_time": total_t,
"token": authResponse["token"]
})
def audio_iter():
conv = miniaudio.convert_frames(format, channels, sample_rate,
source[start:end], device.format,
device.nchannels,
device.sample_rate)
samp_iter = iter(conv)
required_frames = yield b''
old_time = time.time()
while True:
sample_data = bytes(
itertools.islice(samp_iter,
required_frames * channels * field_size))
if not sample_data:
break
new_time = time.time()
old_time = new_time
required_frames = yield sample_data
"""
Convert an audio file to WAV and different sample formats.
"""
import os
import array
import miniaudio
def samples_path(filename):
return os.path.join(os.path.abspath(os.path.dirname(__file__)), 'samples', filename)
src = miniaudio.decode_file(samples_path("music.ogg"), dither=miniaudio.DitherMode.TRIANGLE)
print("Source: ", src)
result = miniaudio.DecodedSoundFile("result", 1, 22050, miniaudio.SampleFormat.UNSIGNED8, array.array('b'))
converted_frames = miniaudio.convert_frames(src.sample_format, src.nchannels, src.sample_rate, src.samples.tobytes(),
result.sample_format, result.nchannels, result.sample_rate)
# note: currently it is not possible to provide a dithermode to convert_frames()
result.num_frames = int(len(converted_frames) / result.nchannels / result.sample_width)
result.samples.frombytes(converted_frames)
miniaudio.wav_write_file("converted.wav", result)
print("Converted sound written to ./converted.wav")
output_info = miniaudio.get_file_info("converted.wav")
print(output_info)