ElevenLabs Speech to Text: Scribe Transcription Guide

ElevenLabs speech to text runs on Scribe v2, a transcription model that supports 90+ languages and costs about 330 credits per minute of audio in the web app, or $0.22 per hour through the API. This guide covers which Scribe model to choose, how far each plan’s credits go, the upload steps and the limits worth knowing before transcribing long recordings.

Last updated: October 2026. Plans and credit costs checked against ElevenLabs’ official pricing page.

What is ElevenLabs speech to text?

ElevenLabs speech to text is the transcription side of the platform, branded Scribe, which turns spoken audio or video into text with word-level timing. It shares the same credit balance as text to speech, so one subscription covers both generating voiceovers and transcribing interviews, podcasts or meetings.

Scribe sits inside a broader audio suite that also includes dubbing, voice isolation and conversational agents. For how it fits with the rest of the platform, see the complete ElevenLabs review.

“a software company that specializes in developing natural-sounding speech synthesis software using deep learning”

Source: Wikipedia, ElevenLabs

ElevenLabs built its name on synthesis, and Scribe is the recognition half of that same audio stack.

Which Scribe model should you pick?

A Scribe model is the speech recognition engine that processes your audio, and ElevenLabs currently offers three active versions. Scribe v2 handles standard file transcription, Scribe v2 Realtime streams text live, and Scribe v2 Medical targets clinical audio, while Scribe v1 is deprecated and should be replaced.

Source: ElevenLabs models documentation, checked October 2026.

For uploaded files, Scribe v2 is the default choice; Realtime only makes sense when text must appear while someone is still talking.

What can Scribe v2 do beyond plain transcription?

Scribe v2 is more than a words-only transcriber: according to ElevenLabs’ speech to text docs, it labels speakers, timestamps each word, tags non-speech sounds and detects the spoken language automatically. These extras turn a raw transcript into something usable for captions, show notes or research coding.

The docs list six features that matter most for creators and developers:

• Speaker diarization: Scribe separates different speakers in one recording, so interview and podcast transcripts arrive already split by voice instead of as one block of text.
• Word-level timestamps: Each word carries precise timing, which makes the output usable for subtitle files, clip searches and syncing text to a video editing timeline.
• Audio event tagging: Non-speech sounds such as laughter or applause are labeled in the transcript, adding context that plain transcription tools usually drop entirely.
• Language detection: Scribe v2 identifies the spoken language automatically across its 90+ supported languages, removing a manual setting step for multilingual content libraries.
• Keyterm prompting: A list of product names, jargon or guest names biases recognition toward the right spelling, billed as an add-on through the API.
• No-verbatim mode: Filler words can be removed during transcription, producing cleaner text for blog posts and show notes without a separate manual editing pass.

Diarization and timestamps are the features that save the most editing time for podcasters and video creators.

How much does ElevenLabs speech to text cost?

ElevenLabs speech to text costs about 330 credits per minute of audio on subscription plans, drawn from the same monthly credit pool as text to speech. That means transcription minutes depend on your plan’s credits, from 10,000 on Free to 600,000 on Pro, and on how much text to speech you generate in the same month.

The table converts each plan’s credits into an approximate number of transcription minutes, assuming the whole balance goes to Scribe.

Source: ElevenLabs pricing page (elevenlabs.io/pricing), checked October 2026. Minutes calculated by GrowthStackKit from the stated ~330 credits per minute.

A weekly one-hour podcast uses about 19,800 credits per episode, or about 79,200 credits a month across four episodes. That fits inside Creator’s 121,000 credits but is far beyond Starter’s 30,000, which shows how quickly the cheaper plan runs out for a regular show.

Creator is the practical floor for anyone transcribing more than a couple of hours a month.

Scribe API pricing for developers

Scribe API pricing is billed per hour of audio rather than in credits, and it is far cheaper per hour than spending subscription credits. ElevenLabs lists Scribe v2 and Scribe v2 Medical at $0.22 per hour, with optional add-ons for keyterm prompting and transcript editing charged on top.

Source: ElevenLabs API pricing page, checked October 2026.

At $0.22 an hour, transcribing a 40-hour archive costs about $8.80 before add-ons, and adding keyterm prompting brings it to about $10.80. Realtime costs roughly 1.8 times as much per hour, so batch processing is the better choice whenever live text is not required. The ElevenLabs API setup guide shows how to get a key and make a first request in Python.

For bulk archives, the API is the cost-efficient route; the web app suits occasional uploads.

How do you transcribe a file in the web app?

Transcribing a file in the ElevenLabs web app takes a handful of steps: open Speech to Text, upload the recording, confirm the settings and download the result. No code is involved, and the cost is deducted from the monthly credit balance at about 330 credits per minute of audio.

1. Open Speech to Text: Signing in and choosing Speech to Text from the dashboard opens the Scribe workspace, where earlier transcripts are also listed for reuse and editing.
2. Upload the recording: Scribe accepts common audio formats such as MP3, WAV and M4A plus video files such as MP4 and MOV, so exported podcast episodes work directly.
3. Set language and speakers: Leaving language detection on suits most files, while enabling speaker separation produces a transcript split by voice for interviews and panels.
4. Add key terms: Entering product names, guest names or jargon before processing improves spelling of words a general model would otherwise guess at incorrectly.
5. Review and export: Reading through the transcript, fixing any names and exporting the text completes the job, ready for captions, show notes or a blog post.

Cleaning noisy audio first with the ElevenLabs Voice Isolator often improves results on outdoor or phone recordings. Five steps, and most of the effort goes into the review pass.

How accurate is ElevenLabs Scribe?

Scribe accuracy is marketed as 98% on the ElevenLabs homepage, but the claim comes without published test conditions, so it describes ideal audio rather than a guarantee. The speech to text docs also note that accuracy varies by language, with some lower-resource languages landing in a weaker error band.

Five habits raise accuracy on real-world recordings:

• Clean source audio: Recordings with a close microphone and little room echo transcribe far better than laptop microphones in large rooms or phone calls on speaker.
• Noise removal first: Running noisy recordings through Voice Isolator before transcription strips background sound that otherwise turns into missing or invented words.
• Keyterm lists: Supplying brand names, acronyms and guest names steers Scribe toward the correct spelling, especially for niche industries and uncommon personal names.
• Separate channels: Recording each speaker on a separate channel lets Scribe process channels independently, which improves attribution in remote interviews with crosstalk.
• Human review pass: A quick read before publishing catches names, numbers and homophones, which remain the most common transcription errors in any automated tool.

Treat the 98% figure as a best case and budget a short human review for anything published.

Realtime transcription with Scribe v2 Realtime

Scribe v2 Realtime is the streaming version of ElevenLabs speech to text, returning text with roughly 150ms latency while audio is still being captured. It is built for live captions, voice interfaces and conversational agents, where waiting for a finished file would break the experience.

Realtime costs $0.39 per hour through the API versus $0.22 for batch Scribe v2, and ElevenLabs’ docs describe tighter keyterm limits in streaming mode. Speech to text is also bundled into ElevenAgents, so builders of voice agents rarely call Realtime separately; the ElevenLabs Agents tutorial covers that path.

Choose Realtime only for live use cases; for recordings, batch Scribe is cheaper and simpler.

Who gets the most from Scribe?

Scribe suits anyone who already pays for ElevenLabs credits and produces spoken content, because transcription runs on the same balance. It is a strong fit for creators turning audio into text assets, and for developers who want cheap per-hour transcription without adding a second vendor.

Five groups get the clearest value:

• Podcasters: Speaker-separated transcripts become show notes, quote cards and blog posts, and the workflow pairs naturally with AI podcast production tools and editing apps.
• YouTubers: Word-level timestamps support captions and searchable transcripts, helping repurpose long videos into shorts, posts and newsletter content with less manual typing.
• Researchers and journalists: Interview transcripts with diarization speed up quote finding and coding, while keyterm prompting helps with specialist vocabulary and unusual source names.
• App developers: Per-hour API pricing and official Python and JavaScript SDKs make Scribe a low-cost transcription layer inside note-taking, meeting or support products.
• Healthcare teams: Scribe v2 Medical targets clinical audio, and HIPAA-related use requires a Business Associate Agreement, which ElevenLabs lists among Enterprise plan inclusions.

Podcasters should also read the guide to making a podcast with AI tools for the full production stack. In short, creators and developers both benefit, for different reasons.

Is Scribe good for captions and subtitles?

Scribe is a solid starting point for captions and subtitles because every word comes with precise timing, which subtitle files depend on. The API returns structured JSON, so developers and editing tools can convert it into caption formats, while creators can review the transcript before publishing.

For content where errors are costly, such as broadcast or accessibility-critical video, ElevenLabs also sells a human-reviewed service called Productions.

“Human-edited transcripts, captions, subtitles, dubs, and audiobooks”

Source: ElevenLabs Productions

For translated subtitles and voice tracks, the ElevenLabs dubbing walkthrough covers localization. Automated Scribe captions work well for most creators, with Productions reserved for high-stakes output.

What limits should you know before uploading?

Scribe limits are mostly about file size, duration and cost rather than features. ElevenLabs’ docs set per-file size and duration caps, split long files into segments for parallel processing, and charge add-ons separately, so a quick check before uploading a long archive prevents surprises.

• Per-file caps: Each upload has a maximum file size and duration, and the docs are inconsistent on multichannel limits, so very long recordings should be checked before processing.
• Credit consumption: At about 330 credits per minute, transcription competes with text to speech for the same monthly pool, which can drain Starter quickly.
• Add-on billing: Keyterm prompting and transcript editing cost extra per hour on the API, so enabling them on every file raises the bill noticeably.
• Language variation: Accuracy differs across the 90+ supported languages, and less common languages may need more correction than English or other widely spoken languages.
• Deprecated Scribe v1: Integrations still calling Scribe v1 should switch to Scribe v2, since ElevenLabs marks the older model as deprecated in its models documentation.

Credit budgeting is the limit most users hit first; the guide to how ElevenLabs credits work explains rollover. Check caps and costs once, then batch your uploads.

Verdict: is ElevenLabs speech to text worth using?

ElevenLabs speech to text is worth using if you already pay for ElevenLabs, because Scribe v2 adds speaker labels, timestamps and 90+ languages on the same credit balance. Heavy transcription users should look at the API, where $0.22 per hour is far cheaper than spending subscription credits.

Scribe is the quiet value in an ElevenLabs plan: transcription you already pay for.

Source: GrowthStackKit, ElevenLabs Review

Start on the Free plan to judge accuracy on your own audio, then pick a plan or the API based on monthly hours.

Frequently asked questions

The 12 most-asked questions about ElevenLabs speech to text.

What is ElevenLabs Scribe?

Scribe is the name of ElevenLabs’ speech to text models. Scribe v2 transcribes uploaded audio and video in 90+ languages, Scribe v2 Realtime streams text with about 150ms latency, and Scribe v2 Medical targets clinical audio. Scribe v1 is deprecated, and ElevenLabs recommends replacing it with Scribe v2 in existing integrations.

How many credits does ElevenLabs speech to text use?

Speech to text uses about 330 credits per minute of audio on subscription plans. The Free plan’s 10,000 monthly credits therefore cover roughly 30 minutes of transcription, Starter’s 30,000 cover about 90 minutes, and Creator’s 121,000 cover about 6 hours, assuming no credits go to text to speech.

How much does the Scribe API cost?

Through the API, Scribe v2 and Scribe v2 Medical cost $0.22 per hour of audio. Keyterm prompting adds $0.05 per hour and transcript editing adds $0.07 per hour. Scribe v2 Realtime costs $0.39 per hour, with add-ons of $0.08 and $0.12 per hour respectively.

How many languages does ElevenLabs speech to text support?

All current Scribe models, including Scribe v2, Scribe v2 Realtime and Scribe v2 Medical, support 90+ languages according to ElevenLabs’ models documentation. Accuracy is not equal across them, though, and the docs note that some less common languages fall into a weaker error band than widely spoken languages.

Is ElevenLabs Scribe accurate?

ElevenLabs’ homepage claims 98% accuracy for Scribe, but no test conditions are published, so treat it as a best case on clean audio. Real results depend on microphone quality, background noise, accents and vocabulary. Removing noise first and adding key terms usually improves transcripts of real-world recordings.

Can Scribe identify different speakers?

Yes. Scribe v2 includes speaker diarization, which separates different speakers in one recording so the transcript arrives split by voice. Recording each speaker on a separate channel further improves attribution, because ElevenLabs can process channels independently and assign speaker IDs by channel.

Does Scribe give timestamps for captions?

Yes. Scribe v2 returns word-level timestamps for each word, which is the timing data subtitle and caption files depend on. Through the API the output arrives as structured JSON, so developers and editing tools can convert it into standard caption formats before publishing.

Can I transcribe video files with ElevenLabs?

Yes. Scribe accepts common video formats such as MP4, MOV, MKV and WebM in addition to audio formats like MP3, WAV and M4A. The audio track is transcribed and charged at the same rate of about 330 credits per minute of audio on subscription plans.

Is Scribe free to use?

Scribe works on the Free plan using the 10,000 monthly credits, which cover about 30 minutes of transcription. The Terms limit free use to non-commercial purposes, so business transcription and client work need a paid plan, starting with Starter at $6 a month.

What is Scribe v2 Realtime used for?

Scribe v2 Realtime is the streaming transcription model, returning text with roughly 150ms latency while audio is still being captured. It suits live captions, voice interfaces and conversational agents. It costs $0.39 per hour through the API, so batch Scribe v2 is cheaper for recorded files.

Can ElevenLabs transcribe medical audio?

Yes. Scribe v2 Medical is specialized for medical and clinical audio and is priced the same as Scribe v2 on the API at $0.22 per hour. HIPAA-related use requires a Business Associate Agreement, which ElevenLabs lists among the inclusions of its custom-priced Enterprise plan.

Should I use the web app or the API for transcription?

Use the web app for occasional uploads where reviewing the transcript on screen matters. Use the API for regular or bulk transcription, because $0.22 per hour of audio is much cheaper than spending about 330 subscription credits per minute, and official Python and JavaScript SDKs make setup straightforward.