mirror of
https://github.com/alphacep/vosk-api.git
synced 2026-02-05 04:49:53 +08:00
4209f3a9fe
* methods get_model_by_name, get_model_by_lang, get_model were added into the model class * importing modules changed to using components; introduced constant MODELS_HOME_DIR; simplified code * added new model folders into init; changed samples and transcriber bin for new mode loader * changed back in cli.py lang arg to args.lang * added 3 directories instead of 1 to check for models * cli.py: added 3 args instead of 1 for model; __init__.py: changed script get_model_path for run get_model_by_name/lang inside current directory * deleted default env var * cli.py: changed arg_name; __init__.py: changed const name, changed model loading only for last directory * deleted unused method * changed by_name, by_lang methods, added download_model method * deleted env variable initialization * deleted print() * deteled unused modules * added progress_bar, added folder AppData/Local/vosk for model search * changed download_model methond; added my_hook method
68 lines
1.6 KiB
Python
Executable File
68 lines
1.6 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
from vosk import Model, KaldiRecognizer, SetLogLevel
|
|
from webvtt import WebVTT, Caption
|
|
import sys
|
|
import os
|
|
import subprocess
|
|
import json
|
|
import textwrap
|
|
|
|
SetLogLevel(-1)
|
|
|
|
sample_rate = 16000
|
|
model = Model(lang="en-us")
|
|
rec = KaldiRecognizer(model, sample_rate)
|
|
rec.SetWords(True)
|
|
|
|
WORDS_PER_LINE = 7
|
|
|
|
|
|
def timeString(seconds):
|
|
minutes = seconds / 60
|
|
seconds = seconds % 60
|
|
hours = int(minutes / 60)
|
|
minutes = int(minutes % 60)
|
|
return '%i:%02i:%06.3f' % (hours, minutes, seconds)
|
|
|
|
|
|
def transcribe():
|
|
command = ['ffmpeg', '-nostdin', '-loglevel', 'quiet', '-i', sys.argv[1],
|
|
'-ar', str(sample_rate), '-ac', '1', '-f', 's16le', '-']
|
|
process = subprocess.Popen(command, stdout=subprocess.PIPE)
|
|
|
|
results = []
|
|
while True:
|
|
data = process.stdout.read(4000)
|
|
if len(data) == 0:
|
|
break
|
|
if rec.AcceptWaveform(data):
|
|
results.append(rec.Result())
|
|
results.append(rec.FinalResult())
|
|
|
|
vtt = WebVTT()
|
|
for i, res in enumerate(results):
|
|
words = json.loads(res).get('result')
|
|
if not words:
|
|
continue
|
|
|
|
start = timeString(words[0]['start'])
|
|
end = timeString(words[-1]['end'])
|
|
content = ' '.join([w['word'] for w in words])
|
|
|
|
caption = Caption(start, end, textwrap.fill(content))
|
|
vtt.captions.append(caption)
|
|
|
|
# save or return webvtt
|
|
if len(sys.argv) > 2:
|
|
vtt.save(sys.argv[2])
|
|
else:
|
|
print(vtt.content)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
if not (1 < len(sys.argv) < 4):
|
|
print(f'Usage: {sys.argv[0]} audiofile [output file]')
|
|
exit(1)
|
|
transcribe()
|