This web-component lets you view the JSON output of Whisper to graphically examine the confidence level and timing of each word. You will want to run whisper with the --word_stamps turned on:
$ whisper --model large-v3 --word_timestamps True media.mp3
You can see a demo at https://edsu.github.io/whisper-transcript
yarn install whisper-transcriptLoad the module:
<script type="module" src="https://cdn.jsdelivr.net/npm/whisper-transcript@latest/dist/index.js"></script>
And use it!
<whisper-transcript audio="media.mp3" url="media.json"></whisper-transcript>If you have a video file that will play natively in the browser you can use the video attribute:
<whisper-transcript video="media.mp4" url="media.json"></whisper-transcript>You will want to clone this repository and then:
yarn install
yarn startThis will open the page in the demo directory. If you like you can swap out the audio.mp3 and audio.json files for ones you've created to see what a transcript you've generated looks like.