Turn Spanish speech into text on your own device: drop an interview, a class, a podcast or a video and get a Spanish transcript, or timed .srt subtitles. Spanish is one of the languages Whisper learned from the most data, so on clear speech even the small in-browser models get most words right, and accents, ñ and the opening ¿ and ¡ are normally written as Spanish is written. Errors cluster where you would expect them: strong regional accents, people talking over each other, music under the voice. Nothing is uploaded, and once the model is cached it works offline.
How it works
- Drop a Spanish recording or video. The spoken language is already set to Spanish, so the model writes Spanish instead of guessing at English.
- Fast (≈41 MB) is usually enough for one clear voice. Pick Accurate (≈77 MB) for strong regional accents, phone calls or a busy room, and choose Subtitles under Output if you need timed captions.
- Press Transcribe, fix any name or term directly in the list, and download .txt for notes or .srt / .vtt for a video.
Questions
- Does it work for Mexican, Argentine or Caribbean Spanish?
- Yes. Whisper has a single Spanish setting for every variety, from Spain to the Southern Cone. Clear, neutral speech gives the cleanest transcript; strong regional accents, slang and very fast speech produce more substitutions. The Accurate model is the steadier of the two on accents, for about twice the download (≈77 MB against ≈41 MB), and it is cached after the first run.
- Does it add punctuation and accent marks?
- Yes. Punctuation, capitals and accents are written by the model itself, the way it learned Spanish is written, including the inverted ¿ and ¡. On mumbled or very fast speech the small models sometimes drop an accent or turn a question into a statement. Play the line, fix it in place, and the correction is in every download.
- Does it show who is speaking?
- No. The transcript is the words in order, without speaker names: telling voices apart is not something this tool does. For an interview, set Output to Subtitles so each phrase becomes its own timestamped line, then add the names by hand; playing a line back makes it quick to hear who said it.
- Does my file get uploaded?
- No. The models and all processing run inside your browser on your own device. You can verify this yourself: open your browser’s DevTools, watch the Network tab, and process a file: there are no uploads. After the first visit most tools also work fully offline.
- How accurate is it?
- The models are OpenAI’s Whisper (tiny and base tiers). On clear speech they are very usable; heavy accents, crosstalk and noisy rooms will show errors. The Accurate tier is noticeably better on difficult audio.
- Which languages work?
- Fifteen, picked in the Language setting: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Turkish, Japanese, Korean, Chinese, Arabic and Hindi. The decoder is conditioned on the one you pick, so it transcribes that language rather than approximating it in English — which also means picking the wrong one returns confident nonsense. The weights understand 99 languages; these are the ones we measured as good enough to offer.
- How long does it take?
- Transcription runs on your processor, so speed depends on the machine: expect around a quarter of the recording’s length, and less on the Fast model than on Accurate.