Simple script for transcribing using Whisper

import os
import json
from transformers import pipeline
import datetime
import signal
from moviepy.editor import AudioFileClip

chunk_length = 30
stride_length = (5, 5)

kwargs = {"language": "english"}

model = "whisper-large-v3-turbo"

print("Initializing pipeline...")
pipe = pipeline("automatic-speech-recognition", model="C:\\Git\\" + model, chunk_length_s=chunk_length, stride_length_s=stride_length, return_timestamps=True, generate_kwargs=kwargs)
#  The original full text can roughly be recovered by doing "".join(chunk["text"] for chunk in output["chunks"]).
print("Pipeline initialized.")

file_extensions = tuple([".wav", ".mp3", ".wma", ".mp4"])

def get_json_file_path(audio_file_path):
  file_extension = os.path.splitext(audio_file_path)[1]
  if file_extension in file_extensions:
    return audio_file_path.replace(file_extension, ".json")

def signal_handler(sig, frame):
    global running
    if (running):
        print("Interrupt registered. Finishing current task.")
        running = False
    else:
        print("Interrupt interrupted, will continue processing.")
        running = True

def seconds_to_hms(seconds):
    hours = (seconds // 3600)
    minutes = (seconds // 60) % 60
    seconds = seconds % 60
    return f"{hours}:{minutes}:{seconds}"

def process_audio_file(audio_file_path):
    """Processes an audio file and saves the result as a JSON file."""
    try:
        # Check if the JSON file already exists
        json_file_path = get_json_file_path(audio_file_path)
        if (json_file_path is None) or os.path.exists(json_file_path):
            return

        audio_clip = AudioFileClip(audio_file_path)
        duration_seconds = int(audio_clip.duration)

        print(f"Processing {audio_file_path}. Duration: "+seconds_to_hms(duration_seconds))
        start_time = datetime.datetime.now()
        print("Started at time: " + start_time.strftime("%I:%M:%S %p"))
        # Process the audio file
        result = pipe(audio_file_path)
        end_time = datetime.datetime.now()
        time_taken = end_time-start_time
        print("Elapsed time: " + seconds_to_hms(time_taken.seconds))
        result["model"] = model
        result["parameters"] = "chunk_length_s=30, stride_length_s=(5, 5)"
        result["source_path"] = audio_file_path

        directory_path = os.path.dirname(audio_file_path)
        last_directory = os.path.basename(directory_path)
        try:
            date = datetime.datetime.fromisoformat(last_directory)
            result["date"] = last_directory
        except ValueError:
            pass

        # Save the result as a JSON file
        with open(json_file_path, "w") as f:
            json.dump(result, f, indent=4)

        print(f"Saved result to {json_file_path}")
    except Exception as e:
        print(f"Error processing {audio_file_path}: {e}")

def process_directory(directory_path):
    """Processes all audio files in a directory and its subdirectories."""

    global running
    if not running:
        return

    for root, dirs, files in os.walk(directory_path):
        for file in files:
            file = file.lower()
            if file.endswith(file_extensions):
                audio_file_path = os.path.join(root, file)
                process_audio_file(audio_file_path)
                if not running:
                    print("Interrupted.")
                    return

running = True
signal.signal(signal.SIGINT, signal_handler)

process_directory("D:\\Recordings")