def ds_process_audio(ds, audio_file, file_handle):
"""Run DeepSpeech inference on each audio file generated after silenceRemoval
and write to file pointed by file_handle
Args:
ds : DeepSpeech Model
audio_file : audio file
file_handle : SRT file handle
"""
global line_count
fin = wave.open(audio_file, 'rb')
fs_orig = fin.getframerate()
desired_sample_rate = ds.sampleRate()
# Check if sampling rate is required rate (16000)
if fs_orig != desired_sample_rate:
print("Warning: original sample rate ({}) is different than {}hz. Resampling might produce erratic speech recognition".format(fs_orig, desired_sample_rate), file=sys.stderr)
audio = convert_samplerate(audio_file, desired_sample_rate)
else:
audio = np.frombuffer(fin.readframes(fin.getnframes()), np.int16)
fin.close()
# Perform inference on audio segment
infered_text = ds.stt(audio)
# File name contains start and end times in seconds. Extract that
limits = audio_file.split("/")[-1][:-4].split("_")[-1].split("-")
if len(infered_text) != 0:
line_count += 1
write_to_file(file_handle, infered_text, line_count, limits)
def ds_process_audio(ds,filepath,file_handle):
w = wave.open(filepath, 'r')
frames = w.getnframes()
fs = w.getframerate()#sampling rate
buffer = w.readframes(frames)
desired_sample_rate = ds.sampleRate()
if fs == desired_sample_rate:
data16 = np.frombuffer(buffer, dtype=np.int16)
else:
data16 = convert_samplerate(filepath,desired_sample_rate)
w.close()
text = ds.stt(data16)#use the deepspeech model to perform speech-to-text
file_handle.write(text+"\n")
def ds_process_audio(ds, audio_file, file_handle, vtt):
"""Run DeepSpeech inference on each audio file generated after silenceRemoval
and write to file pointed by file_handle
Args:
ds : DeepSpeech Model
audio_file : audio file
file_handle : SRT file handle
"""
global line_count
fin = wave.open(audio_file, 'rb')
fs_orig = fin.getframerate()
desired_sample_rate = ds.sampleRate()
# Check if sampling rate is required rate (16000)
# won't be carried out as FFmpeg already converts to 16kHz
if fs_orig != desired_sample_rate:
print("Warning: original sample rate ({}) is different than {}hz. Resampling might produce erratic speech recognition".format(fs_orig, desired_sample_rate), file=sys.stderr)
audio = convert_samplerate(audio_file, desired_sample_rate)
else:
audio = np.frombuffer(fin.readframes(fin.getnframes()), np.int16)
fin.close()
# Perform inference on audio segment
metadata = ds.sttWithMetadata(audio)
infered_text = ''.join([x.text for x in metadata.transcripts[0].tokens])
# File name contains start and end times in seconds. Extract that
limits = audio_file.split(os.sep)[-1][:-4].split("_")[-1].split("-")
# Get time cues for each word
cues = [float(limits[0])] + [x.start_time + float(limits[0]) for x in metadata.transcripts[0].tokens if x.text == " "]
if len(infered_text) != 0:
line_count += 1
write_to_file(file_handle, infered_text, line_count, limits, vtt, cues)