speech → text
Transcribe audio and video to text
Turn an interview, lecture, voice message or video into text. Toola runs OpenAI's Whisper speech model on your own device, so private recordings never leave it. Pick the language or let it detect it, read the text as it appears, correct anything misheard, then copy it or download a TXT file, or SRT and VTT subtitles with timings. It can also translate the speech into English.
Private: your files never leave your device.
Audio or video up to 1 hour. Speech is recognised on your device by OpenAI's Whisper model: nothing is uploaded.
How to use Audio & video to text
Step 1: Add an audio or video file (up to 1 hour).
Step 2: Choose the language and Fast or Best, then press Start.
Step 3: Correct the text if needed, then copy it or download TXT, SRT or VTT.
Getting the most accurate text
Speech recognition is only as good as the recording. A few things make a big difference:
- Choose the spoken language instead of Detect automatically when you know it, especially for short clips.
- Use Best for accents, several speakers, background noise or languages other than English.
- Clean a noisy recording with Remove background noise first, then transcribe the clean file.
- For a long recording you only need part of, cut it with Trim audio first: it saves time.
Whisper adds punctuation and capital letters itself, but names and technical terms are often misheard. Correct them in the text boxes before downloading; the SRT and VTT files use your corrected text.
TXT, SRT or VTT?
| Format | What it is | Use it for |
|---|---|---|
| TXT | Plain text in paragraphs | Notes, articles, emails, AI summaries |
| SRT | Numbered subtitles with times | YouTube, Facebook, Premiere, DaVinci Resolve, VLC |
| VTT | Web subtitles with times | Web players, HTML5 video, Vimeo |
To put the subtitles onto the picture itself, for Instagram, TikTok or WhatsApp, use Add subtitles to video.
Frequently asked questions
Is my recording uploaded?
No. The speech model is downloaded to your browser once and runs on your device. The recording itself never leaves it.
Which languages work?
English, Russian, Ukrainian, Turkish, German, French, Spanish, Italian, Portuguese, Polish and other major languages work well; Best is more accurate for accents and less common languages. Georgian isn't supported yet: in our tests, even the largest Whisper model couldn't recognise Georgian speech.
Fast or Best?
Fast (77 MB) works on any computer and is about 3× faster than real time. Best (about 560 MB, downloaded once) is more accurate but needs a graphics card the browser can use (WebGPU, in Chrome or Edge on most computers).
How long does it take?
With Fast, a 10-minute recording takes roughly 3–4 minutes on a laptop. Best takes about as long as the recording plays. The first time also includes downloading the model.
Can I make subtitles?
Yes. Download SRT or VTT for YouTube, Premiere or VLC, or use Add subtitles to video to write them onto the picture.