Speech to text
Shruti · 22 languages
Shruti, in your browser
Speak or open a recording, and Shruti writes it down. Each model covers its language, English, and the two mixed.
Your audio never leaves your device. The model runs inside this page; the only downloads are the model files, about 32 MB for a language, and after the first load that language also works offline.
Starting…
Where it falls short
- Nine languages (Dogri, Hindi, Kannada, Konkani, Odia, Punjabi, Sindhi, Tamil, Telugu) missed our written comparison with the original model on at least one test set, some by scoring better, some worse. None met every rule. Every number.
- It can get stuck repeating a word or phrase; the text stays as the model wrote it.
- Indian-accented English is often taken for the model's own language. If you speak only English, set "Speech is" to All English.
- Long recordings are cut into 30-second pieces, which can split a word. No punctuation, and numbers come out as words.
- The smallest build needs about 1 GB of memory once loaded; on a phone that struggles, choose "Less memory". Browser memory has not been measured.
- If the microphone is blocked, allow it for this page in your browser's settings, or open a recording instead.