To run the Whisper model locally for audio transcription, you can use the official OpenAI Whisper repository or a community-maintained Python package like faster-whisper (for better performance) or whisper (official). Here’s a step-by-step guide:

1. Install Required Tools

• Python 3.8 or later (recommended: 3.9+)
• FFmpeg (for audio processing)

Install Python

• Download from python.org
• Ensure pip is updated: pip install --upgrade pip

Install FFmpeg

• Windows: Download from FFmpeg’s official site and add to PATH.
• Mac: brew install ffmpeg
• Linux: sudo apt install ffmpeg

2. Install Whisper

Choose one of the following options:

Option A: Official Whisper (OpenAI)

pip install -U openai-whisper

• This installs the official Whisper model.

Option B: Faster-Whisper (Optimized)

pip install faster-whisper

• This is a faster, optimized version of Whisper.

3. Transcribe Audio

Using Official Whisper

whisper audio_file.mp3 --model medium --language English

• Replace audio_file.mp3 with your audio file.
• --model can be tiny, base, small, medium, or large (larger = more accurate but slower).
• --language is optional (e.g., English, French).

Using Faster-Whisper (Python Script)

from faster_whisper import WhisperModel

model_size = "medium"
model = WhisperModel(model_size, device="cpu", compute_type="int8")

segments, info = model.transcribe("audio_file.mp3", beam_size=5)

for segment in segments:
    print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")

• Save this as transcribe.py and run with python transcribe.py.

4. Notes

• GPU Acceleration: If you have an NVIDIA GPU, install faster-whisper with CUDA for faster processing:
  bash

  Copy

  pip install faster-whisper --extra-index-url https://download.pytorch.org/whl/cu118

• Output Formats: Whisper can output .txt, .srt, .vtt, or .json files. Use --output_format to specify.
• Batch Processing: Use a loop in Python to process multiple files.

5. Troubleshooting

• FFmpeg Errors: Ensure FFmpeg is installed and in your PATH.
• CUDA Errors: Update your NVIDIA drivers and PyTorch.
• Memory Issues: Use smaller models (tiny or base) if you run out of RAM.

To extract a segment from 30 to 40 minutes in an MP3 file using FFmpeg, use the following command:

ffmpeg -i input.mp3 -ss 00:30:00 -to 00:40:00 -c copy output.mp3

Explanation of the command:

• -i input.mp3: Specifies the input file.
• -ss 00:30:00: Sets the start time to 30 minutes.
• -to 00:40:00: Sets the end time to 40 minutes.
• -c copy: Copies the audio stream without re-encoding (fast and lossless).
• output.mp3: The output file name.

Notes:

• If you encounter sync issues or errors, try re-encoding the audio:
  bash

  Copy

  ffmpeg -i input.mp3 -ss 00:30:00 -to 00:40:00 -c:a libmp3lame -q:a 2 output.mp3

  • -c:a libmp3lame: Forces MP3 encoding.
  • -q:a 2: Sets audio quality (2 = high quality, range is 0-9).

When using faster-whisper, the Python script you run does not automatically save the transcription to a file—it only prints the results to the console by default. To save the transcription as a text file, you need to modify the script slightly.

How to Save the Transcription to a Text File

1. Update the Script

Here’s a modified version of the script that saves the transcription to a .txt file:

from faster_whisper import WhisperModel

model_size = "medium"
model = WhisperModel(model_size, device="cpu", compute_type="int8")

segments, info = model.transcribe("audio_file.mp3", beam_size=5)

# Open a file to write the transcription
with open("transcription.txt", "w", encoding="utf-8") as f:
    for segment in segments:
        line = f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}\n"
        print(line.strip())  # Print to console
        f.write(line)  # Write to file

print("\nTranscription saved to 'transcription.txt'")

2. Run the Script

Save the script as transcribe.py and run it:

python transcribe.py

3. Check the Output

• The transcription will be printed to the console and saved to transcription.txt in the same directory.

Optional: Save as SRT (Subtitle File)

If you want subtitles (.srt format), use this script:

from faster_whisper import WhisperModel

model_size = "medium"
model = WhisperModel(model_size, device="cpu", compute_type="int8")

segments, info = model.transcribe("audio_file.mp3", beam_size=5)

with open("transcription.srt", "w", encoding="utf-8") as f:
    for i, segment in enumerate(segments, start=1):
        start_time = f"{int(segment.start // 3600):02d}:{int((segment.start % 3600) // 60):02d}:{int(segment.start % 60):02d}"
        end_time = f"{int(segment.end // 3600):02d}:{int((segment.end % 3600) // 60):02d}:{int(segment.end % 60):02d}"
        f.write(f"{i}\n{start_time},000 --> {end_time},000\n{segment.text}\n\n")

print("Transcription saved to 'transcription.srt'")

Troubleshooting

• If you get an error about the file not being found, ensure audio_file.mp3 is in the same directory as the script.
• If the script runs but the output is empty, check if the audio file is valid and contains speech.