Audio & Video to Text — Free Transcription and Subtitle Generator

Most online video transcription and audio services upload your file to their servers, then limit how many minutes you get for free. This tool takes a different route: it downloads Whisper, the open-source speech to text model released by OpenAI, and runs it in your browser with Hugging Face's Transformers.js. Your audio is decoded and transcribed locally. Only the model files are fetched from the internet, and once your browser has cached them, later runs start straight away.

On browsers with WebGPU (recent Chrome and Edge, and Safari on newer devices) the model runs on your graphics card, which is much faster. Everywhere else it runs on the CPU with WebAssembly. That works, but expect it to take a while for long recordings, and keep the tab open until it finishes. The Fast model is a smaller download and quicker; the Accurate model copes better with accents, background noise and languages other than English.

Whisper has a known weakness: given silence or music, it sometimes invents text such as "Thanks for watching!". To avoid that, the tool first scans the audio for stretches that are loud enough to be speech and sends only those to the model, in pieces of up to 28 seconds. Quiet gaps are skipped, a lone filler line in a piece is dropped, and phrases the model repeats over and over are collapsed. The timestamps Whisper returns are then placed back on the file's timeline and long lines are wrapped into subtitles of at most two 42-character lines.

Choose the spoken language before you start; the model doesn't detect it on its own here, and English is assumed by default. The transcript won't be perfect, especially with overlapping voices or heavy accents, so read it through before you publish captions. The editor and translator linked below take the SRT straight from this page.