dotfiles/.local/bin/scripts/vosk

#!/usr/bin/env python3

from vosk import Model, KaldiRecognizer, SetLogLevel
import sys
import json
import os
import wave
import subprocess

SetLogLevel(0)

modelBase = os.getenv('VOSK_MODELS', os.path.join(os.getenv('XDG_DATA_HOME', os.environ['HOME']), 'vosk/models'))
modelPath = os.path.join(modelBase, sys.argv[2] if len(sys.argv) > 2 else 'daanzu')
if not os.path.exists(modelPath):
    print (f"Please download a model from https://alphacephei.com/vosk/models and unpack it to {modelPath}.", file=sys.stderr)
    exit (1)

if len(sys.argv) < 2:
    print (f"Usage: {sys.argv[0]} <audio> [model]", file=sys.stderr)
    exit (1)

sample_rate=16000
model = Model(modelPath)
kaldi = KaldiRecognizer(model, sample_rate)

process = subprocess.Popen(['ffmpeg', '-loglevel', 'quiet', '-i',
                            sys.argv[1],
                            '-ar', str(sample_rate) , '-ac', '1', '-f', 's16le', '-'],
                            stdout=subprocess.PIPE)

def getText(result):
    res = json.loads(result)
    if not 'result' in res:
        return res['text']
    sec = res['result'][0]['start']
    time = "%02d:%02d" % (sec // 60, sec % 60)
    return "%s %s" % (time, res['text'])

while True:
    data = process.stdout.read(4000)
    if len(data) == 0:
        break
    if kaldi.AcceptWaveform(data):
        print(getText(kaldi.Result()))

print(getText(kaldi.FinalResult()))