👁️See🎙️Voice

SeeVoice is a (hacked-together) tool for exploring audio transcriptions from Deepgram.

• Click on a word to play the audio from that point in the transcript
• Red words are low confidence transcriptions
• Hover over a word to see it in the transcript
• With "Details" enabled, see alternate word transcriptions and confidence scores
• Zoom and browse the waveform view
• Currently, only Deepgram transcripts are supported (I have no affiliation)

SeeVoiceDemo.mov

Usage

First, clone the repo.

Then, get a transcript from Deepgram for your audio file:

curl \
  --request POST \
  --header 'Authorization: Token <YOUR_DEEPGRAM_API_KEY>' \
  --url 'https://api.deepgram.com/v1/listen?tier=nova&language=en-US&punctuate=true&diarize=true&smart_format=true&alternatives=4&paragraphs=true' \
  --data-binary @myfile.mp3 \
  -o myfile.mp3.json

Then, move files you want to read into the public folder, or alternatively symlink them. public will be served in webserver root.

ln -s /path/to/myfile.mp3 public/
ln -s /path/to/myfile.mp3.json public/
cp /path/to/myfile2.mp3* public/

Then, run the following commands:

yarn install
yarn dev

This opens a webserver at localhost:5173 by default. Visit this URL in your browser with audioPath and jsonPath query parameters:

http://localhost:5173/?audioPath=myfile.mp3&jsonPath=myfile.mp3.json

Enjoy browsing your transcript!

Tech

SeeVoice uses Vue 3 and Vite, along with a small list of dependencies.

Presently, it runs best in dev mode.