A negative is not a denial
The mark is sealed in at synthesis time, so audio produced by a model before watermarking was enabled on it carries nothing to find. No result here can tell you that we did not generate a clip.
Was this audio made by Speechify? Upload a clip and find out.
A free check for the watermark Speechify seals into the audio its models generate. No account, no sign-in, and nothing about you or your clip is kept.
How to read the answer
A match is evidence. No match is not proof.
The check looks for one thing: an inaudible watermark our synthesis path seals into the audio it generates. Finding it tells you where the audio came from. Not finding it tells you only that the watermark was not recoverable from this file.
Give it a few seconds of speech
Below 3 seconds of clear speech there is not enough signal to read, and the answer is not worth acting on. Music, noise, and silence do not help it.
Editing degrades the mark
Re-encoding at a low bitrate, changing the speed, layering other audio over the top, or clipping a short excerpt out of a longer file all make the mark harder to recover.
From your own code
The same check, as an API.
No key, no account, and no need to come back to this page. The response is the same single verdict.
curl -X POST https://api.speechify.ai/v1/audio/watermark/verify \
-F "audio=@clip.mp3"Building on Speechify? The authenticated endpoint returns the detector's confidence score alongside the verdict.
What we keep
Nothing about you, and nothing about your clip.
Your audio is read to answer the question and is not stored afterwards. This page sets no analytics or advertising cookies and records no profile of the people who use it. There is no form here to fill in and no account to create.
To keep the tool available to everyone, checks are rate-limited per network address and protected by a bot challenge. That is the only thing your address is used for.