Transcribe Video with Whisper

This app uses Azure OpenAI Whisper to transcribe a YouTube video or a local Audio/Video file. You'll need to have a Deployment of Whisper in Azure, which is currently available in the North Central Azure Region.

Prerequisites

• An Azure Subscription
• A Deployment of Whisper (at the time of writing, available in the North Central Azure Region)
• Python 3.11 or higher

Setup

1. Clone this repo
2. Install ffmpeg. If you're on a Mac, you can just brew install ffmpeg
3. Create a virtual environment
4. Install the requirements
5. Use .env.example to create a .env file in the root of the project
6. Run the app

In this example, the following YouTube URL is downloaded, split into 2 audio files, and transcribed: python app.py https://www.youtube.com/watch?v=dQw4w9WgXcQ 2

Notes

Parameters: <YouTube_URL OR Audio/Video_File> [<num_splits>] [<output_file>] [<transcription_file>]

How to use

Transcribe a YouTube Video

The app can be used to directly transcribe a YouTube Video like this:

python app.py https://www.youtube.com/watch?v=dQw4w9WgXcQ 2

Here, the parameters are as follows:

• YouTube_URL: The URL of the YouTube video to transcribe
• num_splits: The number of audio files to split the video into. Defaults to 5

Transcribe a Local Audio/Video File

The app can be used to transcribe a local Audio/Video file like this: python app.py /path/to/local/audio_or_video_file 2

Here, the parameters are as follows:

• Audio/Video_File: The path of the local Audio/Video file to transcribe
• num_splits: The number of audio files to split the video into. Defaults to 5

Transcribe a list of YouTube videos stored in the csv file called youtube_videos.csv

The app can be used to transcribe a list of YouTube videos stored in the csv file called youtube_videos.csv like this:

python batch_processor.py --splits 2

Here, the parameters are as follows:

• splits: The number of audio files to split each video into. Defaults to 10

Also, youtube_videos.csv is the default csv file containing the list of YouTube videos to transcribe.

How the app works

For each YouTube video or Audio/Video file, the app does the following:

• Downloads the YouTube video or processes the local Audio/Video file
• Splits the video into the specified number of audio files
• Transcribes each audio file, using a back-off strategy if the transcription fails due to a timeout
• Writes the transcription to a file in the "output" folder
• Deletes the split audio files and the original audio/video file from the "working" folder

Here:

• YouTube_URL or Audio/Video_File: The URL or Path of the YouTube video or local Audio/Video file to transcribe. If a YouTube URL is provided, it's first downloaded and then split/transcribed. If a local Audio/Video file is provided, it's split/transcribed.
• num_splits: The number of audio files to split the video into. Defaults to 5
• output_file: The name of the output file. Defaults to the code of the YouTube video (e.g., dQw4w9WgXcQ in the example above)
• transcription_file: The name of the transcription file. Defaults to the output_file with a .txt extension (e.g., dQw4w9WgXcQ.txt in the example above)

You can use the batch_processor.py app and the youtube_videos.csv file to process a batch of YouTube videos. The youtube_videos.csv file contains a list of YouTube videos to process, and the batch_processor.py app will process each video in the list.

The app uses the "working" folder to store the downloaded audio/video files and the split audio files during processing. Once the transcription is complete, the split audio files and the original audio/video file are deleted from the "working" folder. The transcribed text files are stored in the "output" folder.