transcribe.py

import os
from pathlib import Path
import nemo.collections.asr as nemo_asr


# def converter(audio_file):
#     converted = audio_file.split(".")[0] + "converted_.wav"
#     cmd_str = f"ffmpeg -y -i {audio_file} -ac 1 -ar 16000 {converted}"
#     os.system(cmd_str)
#     # os.remove(audio_file)
#     return converted
    
    

def transcribe(audio_file):
    # wav_file = converter(audio_file)
    try:
        text = model_kz.transcribe([audio_file])
        return text[0]
    except:
        return 'Try another file format.'
    



language = "kz"

BASE_DIR = Path(__file__).resolve(strict=True).parent

model_kz = nemo_asr.models.EncDecCTCModel.restore_from(restore_path=f"{BASE_DIR}/stt_{language}_quartznet15x5.nemo")