wyreup
export Faster on WebGPU

Transcribe Audio

Convert speech to text using OpenAI Whisper — runs entirely on your device. Pick a model: tiny (~80 MB, fast), base (~250 MB, recommended), or small (~600 MB, best). The chosen model downloads on first use, then works offline.

First run downloads ~238 MB. The model is cached after the first use, then runs offline. Manage downloads on the settings page.
Loading…

About Transcribe Audio

Transcribe Audio converts speech to text using OpenAI's Whisper running entirely on your device — your recording is never uploaded. Pick a model that fits your needs: tiny (~80 MB, fast), base (~250 MB, recommended), or small (~600 MB, best accuracy). The model downloads once on first use and then works offline, so private interviews and confidential calls stay on your machine.

Category
export
Input
Accepts: audio/wav, audio/mpeg, audio/mp4, audio/x-m4a, audio/aac, audio/ogg, audio/webm or audio/flac.
Output
Outputs: text/plain.
Cost
Free, runs in your browser
Memory
high
Install group
speech
Privacy: Transcribe Audio runs entirely on your device. Files you provide never leave your browser — no uploads, no server, no tracking. The page works offline once loaded.

Common uses

  • Turn a recorded interview into a text transcript without sending the audio to a cloud service
  • Caption a voice memo or lecture so you can search and quote it later
  • Draft subtitles from a video's extracted audio track
  • Transcribe a confidential client call where uploading the recording isn't allowed
  • Get a rough text version of a podcast episode for show notes
  • Convert a meeting recording into notes you can edit and summarize

Frequently asked questions

Which audio formats are supported?

WAV, MP3, MP4/M4A, AAC, OGG, WebM, and FLAC. The output is plain text.

Does my audio get uploaded?

No. Whisper runs in your browser on your device. After the model downloads once, transcription happens locally and works offline.

Which model should I choose?

Base (~250 MB) is the recommended balance. Tiny is fastest for quick drafts; small is the most accurate but the largest download.

Why is there a download the first time?

The chosen Whisper model (80–600 MB depending on size) is fetched once and cached, so later runs are fast and offline.

How accurate is it on noisy or accented speech?

It's good for clear speech; for noisy, accented, or difficult audio the hosted Pro version uses the larger Whisper-large-v3 model for better results.

Keywords

  • transcribe
  • speech
  • stt
  • whisper
  • audio
  • subtitles
  • voice

Try next

Enhance Audio
audio

Clean up a rough recording first so the transcript comes out more accurate.
Transcribe Audio
export

Step up to hosted Whisper-large-v3 when the in-browser models struggle with noise or accents.
Word Counter
inspect

Get word and reading-time stats on the finished transcript.
Images → PDF
export

Combine one or more images into a single PDF document.
PDF to Text
export

Extract plain text from a PDF file.
OCR
export

Extract text from images using optical character recognition.
Extract PDF Tables
export

Extract tabular data from a PDF as JSON or CSV. Works well for simple tables with aligned columns and no merged cells. Complex tables, rotated text, and scanned PDFs (images) will not extract cleanly.
Transcribe Audio
export

Hosted Whisper-large-v3 transcription — higher accuracy and better with noisy or accented speech than the free in-browser version. Uses 3 credits per run.
OCR
export

High-quality text extraction from images and scans, powered by a hosted vision model. Output is plain text — chains into translate, summarize, and other text tools. Uses 1 credit per run.
Base64
convert

Encode any file to Base64 text or decode Base64 back to bytes.
URL Encoder
convert

Encode or decode URL components and full URLs.