Free, Unlimited
AI Audio Transcription

It is completely free. There are no limits on duration, number of runs, or character count, and no credit card is required. The only cap is 2 GB per file (500 MB on smartphones); within that, you can use it as often as you like.

No. AI (Whisper or SenseVoice) runs inference inside your browser, so your audio never leaves your device. Nothing is stored on a server and nothing is passed to a third party.

No. Open the page, pick an audio file, and transcription starts. You are never asked for an email address.

Audio: MP3, WAV, M4A, AAC, FLAC, OGG, Opus, AIFF, WMA and AMR. Video: MP4, MOV, MKV, WebM, AVI, WMV, FLV, MPG, MPEG-TS and 3GP. You can drag and drop a file or choose one through the file picker.

Yes. Load an MP4, MOV, MKV, WebM or AVI file and the browser extracts its audio track for transcription. There is no need to convert it to an audio file first. Once transcribed, you can play the video with the result overlaid as captions to check and correct it, or watch it with translated captions. The limit is 2 GB per file (500 MB on phones). Videos saved from X (Twitter) with X Video Downloader can be loaded as they are; to extract just the audio first, use Media Tools.

Speaker diarization is supported (Beta). For recordings with several participants, such as meetings or interviews, each speaker is shown in a different colour. The number of speakers is estimated automatically, so there is normally no need to set it. Speaker labels can be renamed on screen, and speakers that were split by mistake can be merged, and any line can be reassigned or split. When transcription runs on WebGPU in a browser other than Firefox and the speaker count is left on Auto, diarization uses NVIDIA Nemotron 3 Diarization; otherwise it uses pyannote, and in both cases WeSpeaker is used to compare voice characteristics.

Yes — transcripts can be exported as SRT and VTT (WebVTT) subtitles with timecodes, ready to load into video editors or video platforms (most video editors accept SRT). Plain text (TXT), Word (DOCX), PDF, CSV and JSON export and copy to clipboard are also available. To burn the exported SRT directly into the video, use Media Tools (subtitle burn-in).

Yes. Open “Translate” in the result header, choose the source and target languages, and a translation appears under each line. Every export format can carry the original, the translation, or both. If you loaded a video, the translation can also be shown as captions over it during playback. Translation uses the translator built into Chrome 138+ and Edge 148+ on desktop and runs entirely on your device, so the transcript is never sent anywhere. It isn't available on smartphones or in Firefox and Safari. The first use of a language downloads its translation model.

Yes. After the AI model has been downloaded once, transcription works without a network connection. Model sizes range from roughly 40 MB to 5.2 GB depending on which one you choose.

Choosing a model that suits the language improves accuracy. For Japanese, Chinese and Korean, the default SenseVoice Small is light and fast, and for Japanese and Chinese it is at least as accurate as Whisper Large v3 Turbo. For other languages, larger models are more accurate; on devices with WebGPU support, Whisper Large v3 Turbo offers close to the best accuracy while staying fast. On computers with WebGPU you can also choose Qwen3-ASR 1.7B (about 5.2 GB to download the first time), which makes fewer errors than Whisper Large v3 Turbo in Japanese, English and Chinese. Setting the language explicitly instead of leaving it on automatic detection also reduces errors.

Yes, though it runs in a reduced-memory mode (WASM) to protect the device, and files are capped at 500 MB. For long recordings, a desktop computer is recommended.

Only the statistics needed to fix problems. That means your device specs (GPU model, memory, core count), browser and OS, the model and language you chose, your speaker diarization setting (on or off, and the number of speakers you set), the page's display language, how far the run got, the error type and text, the file's format with a rough size and length, and how long it took. The same fields are sent on success too, because a failure rate needs both. Audio, transcripts, file names, and IP addresses are never sent, and the data goes only to this site's own server — never to a third-party service. You can turn it off any time with the switch in the “Help us improve the service” section near the bottom of the page.

Yes. Under “Download” below the result, choose “Word (DOCX)” to save a document that opens in Word. PDF is saved through your browser's print function: choose “PDF”, then save as PDF in the print dialog that opens (on iPhone and iPad, use the share button in the print screen and save to Files). CSV for spreadsheets and JSON for programs are also available. Every format is created inside your browser, so the transcript is never sent anywhere.

Yes. Press “Record with microphone” to record with the microphone of your computer or phone. Press “Stop recording”, then “Start transcription”, and the recording is transcribed. The recording is not sent anywhere either; it is saved in this device's history. You can record for up to 3 hours, or less depending on your device's memory and the selected model; the limit is shown before you start recording. On a phone, locking the screen or switching to another app while recording may stop the recording.