Voiceover Proofer BETA

Your VP of Quality Control.

Checks a voiceover recording against its script and flags missed, added, and changed words, with timestamps. Everything runs on your computer. Neither your audio nor your script is uploaded anywhere.

VP is a second set of ears, not a replacement for yours. For best results, proof your recording by ear first, then run VP to catch anything you missed. The transcriber can mishear or misspell words, especially names and brands, so some flags will be its mistakes, not yours. Gray "Spelling only" flags are the most likely of these. Free to use and provided as is. Donations are optional gifts and never required, but if this app helps you, please consider donating.
Details and Terms of Use

No guarantee

This tool is offered free, "as is," without warranty of any kind. It's a helper, not a replacement for your own proofing. Automatic transcription isn't perfect: it can miss real mistakes, and it can flag words you read correctly because it misheard or misspelled them. You're responsible for checking your recordings before delivery. The author isn't liable for missed errors, lost work, lost clients, or any other loss from using it.

Donations

Donations are voluntary gifts that help support the tool's upkeep. They don't buy a license, support, features, or any guarantee. Everyone gets the same tool whether they donate or not. Donations aren't tax-deductible.

Privacy

Your audio and scripts are processed entirely in your browser and are never uploaded or shared. VP has no accounts, no tracking and no analytics, and it doesn't collect or store anything about you. To work, your browser downloads files from a few services: this page from its host (Netlify), the speech engine from jsDelivr, and the speech model from Hugging Face. Like any website, those services can see basic connection details such as your IP address, but they never receive your audio or script. The donate button takes you to PayPal, which has its own privacy policy.

Credits

Speech recognition uses OpenAI's Whisper model (MIT License), run in the browser with Transformers.js by Hugging Face (Apache 2.0 License). Spelling checks use a word list derived from the SCOWL / Hunspell en_US dictionary by Kevin Atkinson and contributors (see Third-party licenses at the bottom of the page).

1 · Recording

Drop an audio file here or click to choose
WAV, MP3, M4A, FLAC, OGG…

2 · Script

Drop a Word (.docx) or .txt file or click to choose
…or paste the script below

3 · Settings

Transcribing speech means billions of small calculations. A graphics card does thousands of them at once, so it's usually several times faster than your main processor. Leave this off unless proofing fails or gets stuck. Compatibility mode skips the graphics card: slower, but it works on any computer.
Add a recording and a script to get started.
The first run downloads the speech model you picked. After that it's saved in your browser, so later runs start faster. Chrome or Edge recommended.