Veritap docs
 v1.0.0
Get help
● Providers · Audio, video

Speech to text
voice messages as text.

Turns voice messages and the sound of short videos into text, so the message and claim checks can read what was said. Any service with an OpenAI-compatible /audio/transcriptions endpoint works.

OpenAI-compatibleyour own keyper minute

01What it does

For an audio file, a voice message or the sound of a video, the server asks your service for a transcript. The transcript is then redacted on the server (card numbers, account numbers, codes, passwords and ID numbers removed) and checked like a pasted text: the message check reads it, its links are checked, and a transcript over 200 characters also goes to the claim check.

  • The transcript is never a finding by itself and never makes a result look better.
  • When no transcript can be made, "Transcript" is listed as not checked.
  • The message check of a transcript uses one light check from the person's daily allowance, on top of the media units of the recording.
  • The sound of a video is cut out with ffmpeg, so video transcripts need ffmpeg on the server.

02Get a key

Create an account with a speech-to-text service that offers the OpenAI-compatible /audio/transcriptions endpoint (OpenAI itself, or another service that offers the same endpoint), and create an API key there. Note the model id the service uses for transcription.

03Switch it on

  1. Console, Speech to text: turn on Use speech to text.
  2. API base URL: leave empty for https://api.openai.com/v1, or enter your service's address. It must be https on a public address; only the owner can change it, and changing it clears the saved key.
  3. Model: the speech-to-text model id of your service. It is required: without a model the service counts as off.
  4. Paste the API key and press Test connection. It sends a one-second tone made by the server.

With the default address the consent screen names "OpenAI"; with your own address it names "Transcription (via <host>)". See Consent.

04Cost

Billed to your account, per minute of audio, at your service's price (OpenAI: developers.openai.com/api/docs/pricing). Recordings are capped by the console's Longest audio setting (default 300 seconds) and videos by Longest video (60 seconds). The OpenAI-compatible endpoint takes files up to 25 MB, which is why the largest audio cap is 25 MB.