transcribe mp4 video with vosk engine

recognize.py

#!/usr/bin/env python3
# see https://alphacephei.com/vosk

from vosk import Model, KaldiRecognizer, SetLogLevel
import argparse
import wave
import subprocess
import json
from pathlib import Path

SetLogLevel(0)

parser = argparse.ArgumentParser(
    description='извлекает текст из видео',
    epilog = "модели можно скачать с https://alphacephei.com/vosk/models",
)
parser.add_argument("video", type=Path, help='имя файла с видео')
parser.add_argument("model", type=Path, help='каталог с языковой моделью')

args = parser.parse_args()
wavpath = args.video.with_suffix(".wav")
txtpath = args.video.with_suffix(".txt")

sample_rate = 16000
model = Model(str(args.model))

process = subprocess.run([
    'ffmpeg', '-loglevel', 'quiet', '-i', args.video,
    '-ar', str(sample_rate),  # resample
    '-ac', '1',  # single channel
    '-f', 'wav', wavpath
])


wf = wave.open(str(wavpath), "rb")
print(f"wav file with {wf.getnframes()} frames")
recognized = open(txtpath, "w")
rec = KaldiRecognizer(model, sample_rate)

text = ""
firsttime = True
while True:
    data = wf.readframes(wf.getnframes()) # can be varied
    if len(data) == 0:
        break
    if rec.AcceptWaveform(data):
        jres = json.loads(rec.Result())
        if not firsttime:
            recognized.write("\n")
        else:
            firsttime = False
        recognized.write(jres['text'])
jres = json.loads(rec.FinalResult())
recognized.write("\n" + jres['text'])