Transcribana
English transcription
Transcribe English audio or video locally, then review the text with optional English writing checks.
Start transcribingStart transcribing
Best supported on current desktop Chrome. Codec support, speed and memory vary by browser and device. Mobile transcription is not guaranteed.
Drop an audio or video file here
Interviews, meetings, lectures, and everything worth keeping.
Audio and video files are processed locally in your browser.
Supported file formats
MP3, WAV, M4A, AAC, FLAC, OGG, OPUS, MP4, MOV, WebM, MKV, M4V and 3GP with supported audio codecs. Files are checked locally. AVI, WMA and MPEG/MPG program streams are not supported. Codec support varies by browser.
Transcription runs locally in your browser. Your audio and video are not uploaded to our servers.
Advanced settings
Models download on first use. A loaded model is reused; change model or processing mode to release it.
Choose a recording to get started.
Progress will appear here.Your words, ready to use.
Document edits apply to copied text and TXT. Subtitles and JSON preserve the original timed transcription.
Processing details
Timestamps are model estimates. Total RAM and GPU memory are not available to this page; use your browser’s task manager.
Arabic uses a dedicated model with about 1.33 GB of weights and CPU/WASM processing. The first run may take several minutes. General model and GPU choices do not change the Arabic model.
Before you start
Choose English to bypass language detection. Base is the default general model, with Tiny and Small alternatives. WebGPU is preferred when available; CPU/WASM is an alternative with different performance.
A workflow that fits your file
Review accents, technical terms, names and numbers against the recording. Harper can suggest English spelling and grammar changes, but a grammatical sentence can still misrepresent what someone said.
Review and use the result
Accept suggestions individually or keep the speaker’s original phrasing. Copy and TXT use the selected text version. Keep SRT, VTT or JSON when original segment timing is needed.
Review English names, numbers and spoken grammar
An English recording stays in English. Base is the general default; Tiny and Small are alternative multilingual sizes, not promises of a specific accuracy. WebGPU is preferred when available, with a WASM path where practical. Processing time depends on your device and recording.
After recognition, local Harper suggestions can flag writing issues. “He don’t know” might be an error in a report or an intentional quotation in an interview transcript. Compare the audio before accepting it. See the accuracy checklist for names, quantities and negatives that grammar alone cannot verify.
How it works
Choose a recording
Open an audio or video file. We check its audio track in your browser.
Transcribe on your device
Choose the spoken language or Auto Detect. Allow the required model to load.
Review, then export
Listen back to important passages, edit your text and download the format you need.
Your recording never needs to leave your device.
Speech recognition and text checking run locally. The website and model files still need network access. Hosting and model providers can receive ordinary connection information; they do not receive your recording through this tool.
Understand local processingQuestions, answered
Does my recording get uploaded?
No. The browser reads and transcribes it locally. Website assets and model downloads use the network separately.
Can I use it offline?
A model already loaded in an open page can be reused offline. Reloading, switching models or clearing storage may require network access.
Will text corrections change my subtitles?
No. Copy Text and TXT use the selected document version. SRT, VTT and JSON keep the original timed transcription.