ElevenLabs Speech to Text
Transcribes an audio recording into punctuated plain text with ElevenLabs Scribe v1, which detects the spoken language itself and covers more than 90 languages.
Overview
This page runs Scribe v1, which ElevenLabs announced in February 2025 as its first speech-to-text model. You give it a recording and get back a transcript with punctuation and capital letters. ElevenLabs lists support for more than 90 languages. The page does not ask which one is spoken; Scribe works that out from the audio.
ElevenLabs reports that Scribe beat Gemini 2.0 Flash, Whisper Large V3 and Deepgram Nova-3 in its FLEURS and Common Voice tests. The launch post singles out Serbian, Cantonese and Malayalam, where it says competing models often exceed a 40% word error rate, as languages where Scribe makes far fewer errors. Both claims come from ElevenLabs' own testing.
What you can build
- A text archive of podcasts, lectures or recorded calls that people can search instead of scrubbing through audio.
- Meeting and voice-memo notes, where the transcript goes to a language model that writes the summary or the action items.
- Quotes and article drafts pulled from a recorded interview.
- The first step of a translation flow: write the speech down in its original language, then translate the text.
- Moderation or quality review of recorded speech, where staff read transcripts instead of listening to every file.
Inputs
One recording per run. Upload an MP3, WAV, M4A or WebM file, or paste a link to it in the URL field. The link must be publicly reachable.
The form has no other fields. For speech inside a video, extract the audio track to one of those formats first.
Outputs
A .txt file with the transcript as one unbroken paragraph. It carries no timestamps and no speaker names.
The first sample comes from a short WAV recording of six English test sentences, each returned with its own capital letter and full stop. The second comes from the MP3 that the form loads by default, which is a passage of verse read aloud: the transcript ends most lines with a comma or semicolon and spells the old contractions 'Neath and 'Tis with their apostrophes.
Limitations
- Text is all you get. Scribe can return word timings and speaker labels, but this page drops them, so it does not suit subtitle files or transcripts that need to show who said what.
- It works on finished recordings only. There is no live or streaming mode on this page.
- You cannot tell it which language to expect. Detection is always automatic on this page.
- ElevenLabs now lists Scribe v1 among its deprecated models and names Scribe v2 as its replacement.
Safety & compliance
If a recording contains other people's voices, get their consent before you transcribe it.
API quick start
Run speech-to-text with a single API call.
{
"inputAudio": "https://your-cdn.com/input.mp3"
}curl -X POST "https://api.wiro.ai/v1/Run/elevenlabs/speech-to-text" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"inputAudio": "https://your-cdn.com/input.mp3"
}
JSONOther models you might like
Explore similar audio & speech models.



