Free Audio to Text

Worked example

A 30-minute interview saved as MP3 takes about 27 MB. With the "Better" model (an 81 MB download the first time, then cached) a typical laptop needs roughly half an hour, because the tool processes about one second of audio per second plus about 45 seconds to start. The "Faster" model (45 MB) takes about half that. The estimate is shown before you start and corrects itself after the first window. The result can be saved as SRT, VTT, ASS, plain text or LRC.

Which settings to use, and the limits

• Use Better unless the device is short on memory. "Better" is about 81 MB to download and "Faster" about 45 MB. Both are kept in your browser after the first use. On a phone, or if the run fails, try "Faster".
• Set the spoken language when detection is unsure. "Detect automatically" works well for clear speech. If the recording starts with music or switches languages, pick the language under "Spoken language". "Translate to English" gives English text from speech in another language.
• Mind the length limits. The tool warns when a recording passes 2 hours on a computer or 30 minutes on a phone. Past 5 hours on a computer, or 90 minutes on a phone, only the first part is transcribed and the page says so. Cut the rest into a separate file to do it next.
• Stop at any time and keep the text. "Stop and keep what is done" ends the run and leaves everything transcribed so far to copy or download. On a phone, keep the tab open and the screen on, since the browser may pause a background tab.
• Edit before you export. Click a time to hear that moment, fix the wording in place, or use Find and Replace. Copy and every download use your edits. Several files dropped together are transcribed one after another and saved as a ZIP.

Common problems and how to fix them

It says the file changed or moved after I picked it.

The browser lost access to the file, usually because it was moved, renamed or still syncing from cloud storage. Choose the file again. A file copied to the local disk first is the most reliable.

The file could not be read, so nothing was transcribed.

No model is downloaded in this case. If the file is a video, save its sound as an MP3 with the Audio Extractor and transcribe that. If it is audio, re-export it as MP3 or WAV and try again.

The model download stopped or never finished.

A dropped connection or a closed tab interrupts the download. Check your connection and press "Transcribe" again. Once the model has downloaded it is cached, and later runs do not need it again.

The transcription did not finish, or stalled part-way.

This is usually the browser running out of memory or the tab being put to sleep. Try the "Faster" model, close other tabs, and keep this tab in front. If it stopped part-way, the transcript so far is shown and can be downloaded.

No speech was found in the recording.

The audio may be silent, mostly music, or very quiet. Play the original to check. If speech is faint, raise it first with the Audio Normalizer and transcribe the result. A file with no audio track cannot be transcribed.

How it compares

Transcribing in the browser means the recording never leaves your device, which matters for interviews, medical notes and legal recordings. The cost is speed and setup: the first run downloads a model, and a phone is several times slower than a laptop. The model is a small Whisper model, so it is less accurate than the largest ones on heavy accents, background noise or technical vocabulary. Plan on a quick read-through and corrections in the editor.