TL;DR / Key Takeaways
- Scribe v2 is ElevenLabs' batch transcription model; Scribe v2 Realtime handles live audio in about 150ms.
- API price: $0.22 per audio hour for Scribe v2, $0.39 for Realtime, on every plan.
- Features: 90+ languages, speaker labels for up to 32 speakers, word-level timestamps, keyterm prompting, audio event tags.
- Files up to 3GB and 10 hours; MP3, WAV, M4A, MP4, MOV and many other formats accepted.
- In the web app, transcription uses 330 credits a minute from your plan.
Short answer: ElevenLabs converts audio and video to text with its Scribe models. Scribe v2 handles uploaded files, including MP3, in 90+ languages, labels up to 32 speakers and adds word-level timestamps. Through the API it costs $0.22 per hour of audio ($0.39 for Scribe v2 Realtime), and in the web app it uses 330 credits a minute from your plan. Files can be up to 3GB and 10 hours long.
How we checked: features, limits and prices come from the ElevenLabs speech to text page, API pricing page and speech to text documentation, all checked on elevenlabs.io on November 1, 2026. We did not run our own accuracy test. Some links are affiliate links: if you sign up, ElevenLabs may pay us a commission at no cost to you.
The Scribe models
| Model | Use | API price per audio hour | Notes |
|---|---|---|---|
| Scribe v2 | Batch transcription of files | $0.22 | 90+ languages, diarization, timestamps, keyterms, entity detection |
| Scribe v2 Realtime | Live audio, agents, meetings | $0.39 | About 150ms latency, word-level timestamps, voice activity detection |
| Scribe v2 Medical | Clinical audio | $0.22 | HIPAA-eligible with BAAs and Zero Retention Mode |
Add-ons are billed per hour too: keyterm prompting costs $0.05 an hour on Scribe v2 and $0.08 on Realtime, and transcript editing with natural-language instructions costs $0.07 and $0.12. Silence counts toward billed time, so trimming long gaps before upload saves money.
One inconsistency to flag: the FAQ on ElevenLabs' speech-to-text page still says Scribe starts "from $0.40 per hour", while the API pricing table and its FAQ both say $0.22. We use the pricing table's figure; check the live page before budgeting a large job.
What Scribe v2 can do
- 90+ languages, with smart language detection.
- Speaker diarization for up to 32 speakers, so interviews and meetings come back labelled.
- Word-level timestamps for captions and subtitles.
- Keyterm prompting with up to 1,000 words or phrases, useful for names, products and jargon.
- Dynamic audio tagging of non-speech events such as laughter or footsteps.
- Entity detection and redaction of sensitive information through the API.
- Multichannel mode for up to 5 channels, each assigned its own speaker.
- Webhooks so long files can be processed asynchronously.
How to turn an MP3 or video into text
- In the app: sign in to ElevenLabs, open Transcription (speech to text), upload your MP3, WAV, MP4 or other file, pick the language or leave it on auto-detect, and download the transcript or export subtitles.
- Through the API: create an API key, then send the file to the Speech to Text endpoint with
model_idset toscribe_v2. The response includes the text, detected language, words with timestamps and a speaker ID for each word.
Accepted audio formats are AAC, AIFF, OGG, MP3, OPUS, WAV, FLAC, M4A and WebM. Video formats are MP4, AVI, MKV, MOV, WMV, FLV, WebM, MPEG and 3GPP.
Limits
| Limit | Value |
|---|---|
| Max file size | 3GB |
| Max duration (standard mode) | 10 hours |
| Speakers labelled | Up to 32 |
| Multichannel | Up to 5 channels |
| Keyterms | Up to 1,000 |
| Concurrent transcriptions | 8 Free, 12 Starter, 20 Creator, 40 Pro, 60 Scale and Business |
How much transcription each plan includes
| Plan | Price | In the app (330 credits a minute) | Through the API (Scribe v2) |
|---|---|---|---|
| Free | $0 | 12 minutes (as listed) | about 4.5 hours |
| Starter | $6 | about 1 hour 31 minutes | 27 hours |
| Creator | $22 | about 6 hours 7 minutes | 100 hours |
| Pro | $99 | about 30 hours 18 minutes | 450 hours |
| Scale | $299 | about 90 hours | 1,359 hours |
| Business | $990 | about 303 hours | 4,500 hours |
The gap between the two columns is real: the API prices transcription at $0.22 an hour, while the app spends the same credits you use for speech. If you transcribe a lot, calling the API (or a tool built on it) gives far more hours per dollar. In the app, credits also cover text to speech, so heavy transcription leaves less for voiceovers.
Accuracy: what ElevenLabs and others say
ElevenLabs claims over 98% transcription accuracy for Scribe v2 and publishes charts comparing it favourably with OpenAI, Google and Deepgram models. It also publishes expected word error rates by language, from 5% or lower for its top tier down to 20% to 50% for the weakest languages. Independent reviewers are more cautious: DiscoverAI argues accuracy depends on your speakers, noise and terminology, and recommends benchmarking on your own audio. Telenow notes that Scribe is cloud-only, with no self-hosted or air-gapped option.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Who it suits
- Podcasters and video creators who want captions and transcripts in the same tool as their voiceovers.
- Developers building voice agents or meeting tools that need realtime transcription.
- Teams transcribing many languages, or clinical audio with the Medical model and a BAA.
If you mainly want to edit a recording by editing its transcript, Descript is built around that workflow. For everything else, Scribe is worth a try on the Free plan before you commit.
Frequently asked questions
Can ElevenLabs convert MP3 to text?
Yes. Scribe accepts MP3 along with AAC, AIFF, OGG, OPUS, WAV, FLAC, M4A and WebM audio, and MP4, MOV, MKV, AVI and other video files. Upload in the web app or send the file to the Speech to Text API.
How much does ElevenLabs speech to text cost?
Through the API, Scribe v2 costs $0.22 per audio hour and Scribe v2 Realtime $0.39. Keyterm prompting adds $0.05 an hour ($0.08 Realtime). In the app, transcription uses 330 credits per minute from your plan.
Is ElevenLabs speech to text free?
The Free plan includes some transcription: the speech-to-text page lists 12 minutes a month in the app, and the API pricing page lists about 4.5 hours of Scribe v2 for Free API usage. Free-plan output cannot be used commercially.
Does ElevenLabs Scribe identify different speakers?
Yes. Scribe v2 labels up to 32 speakers (speaker diarization) and returns word-level timestamps. Multichannel mode can also assign speakers by audio channel, up to 5 channels.
How many languages does ElevenLabs Scribe support?
ElevenLabs says 90+ languages, grouped by expected error rate. English, Spanish, French, German, Japanese, Italian and Portuguese are among those in its best tier (5% word error rate or lower).
What is the maximum file size for ElevenLabs transcription?
3GB per file and up to 10 hours of audio in standard mode, per ElevenLabs' documentation.