whisper.py

whisper.py

from faster_whisper import WhisperModel
from pathlib import Path
# import opencc
import argparse

# python3 .\whisper.py '.\audio.mp3' --language zh --keywords 這是說正體中文的節目 --model turbo

# parse command line arguments
parser = argparse.ArgumentParser(
    description="Transcribe an audio file using Whisper and save as SRT/VTT."
)
parser.add_argument("filename", type=str, help="Path to the audio file")
parser.add_argument("--model", type=str, help="Model to use", default="large-v3")
parser.add_argument(
    "--language", type=str, help="Language of the audio file", default="zh"
)
parser.add_argument(
    "--keywords", type=str, help="Initial prompt for the model", default=""
)
parser.add_argument(
    "--offset", type=int, help="Offset for the timecode (ms)", default=0
)
args = parser.parse_args()
print(args)

# Load Whisper model
model = WhisperModel(args.model, device="cpu", compute_type="int8")
filename = args.filename

# Transcribe the audio file
initial_prompt = (
    f"{args.keywords}"
)

segments, info = model.transcribe(
    filename,
    language=args.language,
    initial_prompt=initial_prompt,
    word_timestamps=True,
    vad_filter=True,
    vad_parameters={"min_silence_duration_ms": 500, "max_speech_duration_s": 10},
)


# Function to format time in SRT format with more precise milliseconds
def format_timestamp(seconds):
    seconds += args.offset / 1000
    milliseconds = int(seconds * 1000) % 1000
    seconds = int(seconds)
    minutes, seconds = divmod(seconds, 60)
    hours, minutes = divmod(minutes, 60)
    return f"{hours:02d}:{minutes:02d}:{seconds:02d},{milliseconds:03d}"

# Save VTT
file_path = Path(filename + ".vtt")
with open(file_path, "w", encoding="utf-8") as f:

    print("WEBVTT\n", file=f)
    for segment in segments:
        start = format_timestamp(segment.start)
        end = format_timestamp(segment.end)
        text = segment.text.strip().replace("-->", "->")
        print(f"{start} --> {end}\n{text}\n", file=f, flush=True)
        print(f"{start} --> {end}\n{text}\n", flush=True)

print(f"VTT file saved to {file_path}")