Speech to Text (Whisper)

Whisper speech recognition that runs on your device

This tool turns spoken words into text with Whisper, the open speech recognition model released by OpenAI. Unlike most online transcription services, nothing is sent to a server: the audio is decoded by your browser, resampled to 16 kHz mono and passed to the model in a background Web Worker. The page picks the fastest engine available — WebGPU on graphics cards and browsers that support it, or WebAssembly on the CPU everywhere else. The first run downloads the model weights from Hugging Face; after that they stay in your browser cache, so later sessions start much faster.

You can transcribe interviews, lectures, podcasts, voice memos and meeting recordings, or the audio track of a video file. Short clips can be recorded straight from your microphone. Long recordings are processed in overlapping 30-second windows, and the progress bar shows how many windows are finished.

Transcripts, translations and subtitles

Choose Transcribe to get text in the original language or Translate to English to let Whisper write an English version of speech in another language. Auto-detect listens to the first 30 seconds to identify the language; if your recording starts with music or silence, selecting the language yourself gives better results. Turn on timestamps to see when each sentence was spoken, and download an SRT subtitle file that you can load into video editors, YouTube or media players.

Tips for accurate results

Clear audio matters more than anything else. Record close to the microphone, avoid background music and let one person speak at a time. Whisper base is a compact model, so names, jargon and heavy accents may need a quick manual correction — always proofread before publishing. If your device has little memory, close other heavy tabs before transcribing long files, and keep this tab open until the transcript is complete.