Transcribe pregações peregrinos after downloading from soundcloud

whispertranscribe.py

# pip install scdl
# this downloads all tracks from that user to the path specified
# -a download all
# -c skip and continue those already downloaded 
# scdl -l https://soundcloud.com/ipperegrinos -a -c --path /home/andre/music/ipperegrinos
%cd /mnt/Data/ipperegrinos

import subprocess
import pathlib 
from pathlib import Path
import whisper

def process_audio(input_file: Path, output_file: Path):
    """
    Process the audio file by applying loudness normalization, noise reduction,
    and a high-pass filter using FFmpeg. Then convert to WAV format for Whisper.
    """
    # Step 1: Apply audio processing with FFmpeg and save as MP3
    ffmpeg_command = [
        'ffmpeg', '-i', str(input_file),
        '-vn',  # Exclude the video stream (cover art)
        '-af', 'loudnorm, afftdn=nf=-25, highpass=f=150', 
        '-ar', '16000',  # Set audio sample rate to 16kHz - microphone maximum 16kHz
        '-ac', '1',  # Set audio channels to mono
        '-b:a', '64k', # enough bit rate
        str(output_file)
    ]   
    # Run FFmpeg to process MP3
    try:
        subprocess.run(ffmpeg_command, check=True)
    except subprocess.CalledProcessError as e:
        print(f"Error processing {input_file}: {e}")
        return    

def save_transcription(model, audiopath, txtpath):
    # Transcribe the audio not using without_timestamps=True - 
    # default behavoir is better in quality?? - lets remove the stamps after. 
    result = model.transcribe(str(audiopath.absolute()), language="pt", without_timestamps=True) 
    transcription = result["text"]
    with txtpath.open('w', encoding="utf-8") as f:
        f.write(transcription)
    return result 

# Load the Whisper model
model = whisper.load_model("medium", download_root=str(pathlib.Path.cwd()))  

for path in pathlib.Path.cwd().glob('*.mp3'):    
    propath = path.parent / 'processed' / path.name
    txtpath = path.parent / 'text' / (path.stem + '.txt')      
    if not propath.exists():
        process_audio(path, propath)    
    if not txtpath.exists():
        tr = save_transcription(model, propath, txtpath)