Find Any Word in Any Recording
Upload a recording — or a whole folder of voice messages — type a word or phrase, and jump straight to every moment it was said. No scrubbing through an hour of audio. 100% in your browser.
Drop an audio or video file
or click to browse. Best with files under 30 min on most browsers. Cap at 60 min — split longer files first with our Audio Splitter.
Don't have a file? Record one with our voice recorder to test how transcription works.
100% in your browser. Audio stays on your device. The Whisper AI model downloads once (~40 MB) from our servers, then runs locally for every transcription. We can't access your audio because it never leaves your computer. Privacy policy.
Runs free in your browser. Keep this tab open while it runs — we'll chime if you switch tabs. Models cache after first download. Need translation? Use the dedicated Audio Translator.
Transcript
Find words in audio — search any recording, free and private
Audio Word Finder lets you search a recording for any word or phrase instead of scrubbing through it minute by minute. It uses OpenAI's open-source Whisper speech-recognition model running entirely in your browser via WebAssembly: drop in your audio, it transcribes on-device, then you type what you're looking for and jump straight to every moment it was said. The first time you transcribe, your browser downloads a ~40 MB model file from our servers; after that, everything runs locally. Your file never gets uploaded to any server — not ours, not OpenAI's, not anyone's.
Need a plain transcript or subtitle file instead of a word search? Use Audio Transcription or the SRT Generator.
What it's good for
- Finding a quote, name, or term in a long interview, podcast, lecture, or meeting recording — type it once, jump to every mention.
- Searching a whole folder of WhatsApp voice notes or voicemails at once — upload them together and see which message each result came from.
- Spotting how often a word comes up with the built-in word cloud of the most-said words.
- Privacy-sensitive audio you don't want on a third-party server — confidential interviews, internal meetings, personal voice messages.
How the word search works
Once a recording is transcribed, type a word or phrase into the search box — you can even comma-separate several words to hunt for any of them at once. Every match is highlighted in the transcript, and each chunk has a play button that seeks the audio straight to that moment, so you never have to scrub. The transcript is fully editable too: hit ✎ Edit to fix a misheard word, or use Replace all to correct a name or term everywhere at once before searching. Matching is whole-word, so searching "a" won't light up every letter a inside other words.
What it's not so good for
- Heavy background noise, music behind voice, or multiple overlapping speakers — the in-browser model struggles with these, which can hurt search accuracy.
- Heavy accents or non-mainstream dialects — bigger Whisper models handle these better but are too heavy for a browser.
- Knowing who said what (speaker labels) — not supported.
- Recordings longer than about 60 minutes total — we cap input length to keep browser RAM under control; search longer material in batches.
Works in 99 languages
Whisper recognises speech in 99 languages, so you can search a recording for a word whatever language it's spoken in — a Spanish podcast, a Mandarin interview, a French lecture. Pick the language (or let it auto-detect) and the transcription, search, and word cloud all work the same way.