What audio formats are supported for transcription?

We support all major audio formats including MP3, WAV, M4A, AAC, and FLAC. Upload directly from your device or cloud storage. No conversion required.

How fast is the transcription process?

Our AI processes audio files in seconds, even long recordings. With Scribe, you get high-accuracy, speaker-labeled transcripts really fast.

Can I edit the transcript after it's generated?

Every transcript opens in an editor built for cleanup: click a word to fix it, adjust where segments start and end, and correct any speaker label Scribe got wrong. Because each word carries its own timestamp, your edits stay aligned with the audio, and the exported file reflects every change.

What makes these transcripts better than other tools?

Scribe generates a structured AI transcript. Every transcript arrives with up to 32 speakers labeled, every word timestamped, and non-speech sounds like laughter and applause tagged, across 90+ languages. That structure makes a text file searchable and quotable: jump to the exact second a phrase was said and know who said it.

What export options are available?

Seven formats: TXT, DOCX, PDF, JSON, SRT, VTT, and HTML. Pick TXT or DOCX for notes and articles, SRT or VTT when the audio pairs with video captions, and JSON when a developer needs the timing data. Every export keeps the speaker labels and timestamps from your transcript.

Transcribe audio to text with AI

Use ElevenLabs's audio to text converter to turn interviews, lectures, and voice memos into accurate, speaker-labeled text, even with background noise, heavy accents, or across hours of tape. Try it today in 90+ languages.

Convert Audio to Text

Trusted by 1M+ users · Free to start

InterviewsClear speech, even with bad audio

PodcastsEvery speaker labeled, ready to edit

LecturesHours of audio, transcribed in minutes

LyricsCatches every word through the mix

CallsHandles thick accents without breaking

Interviews.pdf

Not just transcription. Audio understanding

ElevenLabs Audio to Text identifies who's speaking, when they're speaking, and what's happening around them - delivering structured, actionable transcripts every time.

#1 Accuracy

Scribe outperforms every major competing ASR model in benchmark tests. Even with distant microphones, strong accents, and low-quality phone recordings, Scribe provides an industry-leading word error rate.

Edit the transcripts

Click a word to correct it, split or merge segments, and reassign a mislabeled speaker without leaving the page. Word-level timing keeps every edit anchored to the audio.

Amidst the outer atmosphere of the planet Aurora, the sky shimmered with fractured light, as though the planet's veil were made of stained glass suspended in space.

Sensors pulsed with irregular patterns, the kind no algorithm could quite reconcile.

Amidst the outer atmosphere of the planet Aurora, the sky shimmered with fractured light, as though the planet's veil were made of stained glass suspended in space.

90+ Languages and accents

Scribe transcribes 90+ languages, including widely underserved ones. It can also automatically detect languages for you, providing precise audio to text AI transcription. Even interviews that drift between languages come back as one coherent transcript.

Japanese

Hindi

Polish

Swedish

Mandarin

Vietnamese

French

Wide variety of formats

Upload MP3, WAV, M4A, FLAC, OGG, or even video files, and download the result as TXT, DOCX, PDF, SRT, VTT, JSON, or HTML. One tool covers every device you record on.

Audio Event Tagging

Scribe marks non-speech events such as laughter and applause, so a lecture transcript shows where the room reacted in real time.

Speaker Timestamps

Scribe labels up to 32 speakers and timestamps each word, so you always know who said what, and exactly when, in a panel or group interview.

From audio to text in three simple steps

Upload your audio

Drag in a file from your device or cloud storage. We accept MP3, WAV, M4A, AAC, FLAC, and OGG, plus every major video format, so nothing needs converting first.

Scribe processes it

Scribe identifies each speaker, timestamps every word, and holds its accuracy through crosstalk and room noise. Recordings over 8 minutes are split and processed in parallel, so a long file does not mean a long wait.

Download clean, structured text

Read the transcript with speaker labels and audio event tags already in place, correct anything by clicking the word, and export in the format your work needs.

Millions of words transcribed, and counting

“I use ElevenLabs primarily for transcribing audio messages, and I find its accuracy to be a major highlight. This precision allows me to analyze students' reading fluency effectively, even when the speaker is a young student still learning to read, which is crucial for understanding each student's progress.”
Pedro A.
Head of technology
“Perfect for transcribing interviews - and the voice quality is amazing when preparing for a speech.”
Izabela M.
Customer Experience Researcher
“Remarkable inference speed of the Scribe v2 model by ElevenLabs, delivering near real-time latency on transcription requests, significantly faster than other models we've tried.”
Vedaswaroop I.
Founder