SubGrab

"No Speech Was Detected": What It Means and Why You're Refunded

When SubGrab says no speech was detected, the audio held nothing the speech model could transcribe. Why it happens, why you're refunded, and what to check first.

Short answer

"No speech was detected" means the audio held nothing the speech model could transcribe: music, silence, background sound or effects. It's also the answer when the AI writes only the stock lines it invents over music and silence, such as "Thank you.". Nothing is charged for it: the credit is refunded automatically.

You paste a link, the transcription runs, and instead of a transcript the page says: "No speech was detected in this video." It means the audio held nothing the speech model could turn into words. You aren't charged for it. Here's what causes it, and what to check before trying again.

What the Message Means

SubGrab transcribes speech: people talking, a narrator, a presenter, singing with words. When the audio is music, silence, applause or sound effects, there's nothing to write down, so instead of an empty page you get this message and your credit back.

Since November 2026 the same message also covers a quieter case. AI speech models don't always stay silent over music. They often write a stock phrase instead: "Thank you.", "Music", a subtitle credit. When a whole transcript is nothing but those, there was no speech to transcribe, so it's treated exactly like an empty one.

The Common Causes

CauseWhat it sounds likeWhat to do
Instrumental music or a soundtrackMusic with no wordsNothing to transcribe; try a video with speech
A silent screen recording or slideshowSilence, maybe clicksNothing to transcribe
The link leads to a teaser or clipMusic, then it endsFind the link to the full video
Speech buried under loud musicWords you can barely make outA cleaner copy of the audio may help
No sound at all in the fileNothing, the player shows no audioThat's a different message: "This video doesn't have an audio track"

Why "Thank You." Counts as No Speech

Speech models learned from huge amounts of video paired with subtitles, and subtitles are full of text that isn't speech: "Thank you for watching" over an outro, "[Music]" over a jingle, a subtitler's credit at the end. Fed music or silence, the model writes what it most often saw there (why your transcript says "Thank you" when nobody did).

Researchers measured it. Running Whisper large-v3 on 301,317 clips with no speech, "thank you" made up 24.76% of what it invented and "thanks for watching" another 10.32% (Baranski et al., ICASSP 2025).

On SubGrab, in the month to 17 November 2026, about 20 of 2,670 AI transcripts were nothing but such lines: a single ".", "Thank you." repeated, "Music", the Russian for "musical intro". Each of those users had paid for a transcript of nothing. Now those count as no speech, and the credit comes back.

You're Not Charged

A transcription that finds no speech is refunded automatically, including the speaker-labels add-on if you turned it on. The page says so under the message. Nothing needs to be requested.

What to Check Before Trying Again

  • Play the video yourself. Is there talking where you expected it, at normal volume?
  • Check the link. Some links lead to a preview, a trailer or one clip of a longer post. Open the full video and copy its own link. Every link format that works lists the shapes per platform.
  • Look for the full episode. A highlight reel set to music has no speech, even when the full broadcast does.

If someone plainly speaks in the video and the message still appears, write to help@subgrab.com with the link, and we'll look at what the audio held.

Frequently asked

Was I charged for it?

No. When a transcription finds no speech, the credit taken for it is refunded automatically, and the page says so. If you turned on speaker labels, that add-on is refunded too. There's nothing to request, and nothing is left owing on your account.

Why does a music video count as no speech?

Sung lyrics often transcribe well, but instrumental music, a backing track or an intro without words gives the model nothing to write. Speech models tend to fill that gap with stock phrases such as "Thank you.", so a transcript made only of those is treated as no speech and refunded.

People talk in the video. Why does it say no speech?

Check that the talking is in what the link points to: a link can lead to a teaser, a short clip or another version of the video. Very quiet or heavily muffled speech can also be missed. If speech plays clearly and it still says none, write to help@subgrab.com with the link.

Does one stray "Thank you." mean something went wrong?

No. A line like that can appear over a pause inside real speech, and the speech around it is transcribed normally. SubGrab treats a transcript as no speech only when every line is one of those stock phrases. A single real sentence keeps the whole transcript, exactly as it was written.

Written by the SubGrab team. About SubGrab says who we are and where your audio goes; questions and corrections go to help@subgrab.com.