The app, function
Text from photos and sound: OCR and speech to text, on the phone, with no internet
In Mynd, the text in a photo or a PDF is read by OCR, and the speech in a voice message, a recording or a video becomes written text with STT, speech to text. It all happens on the phone, with no internet: the photo and the recording never leave it. The text stays in the record, goes into search, and can be copied and sent with the rest.
Start version. The app screens are in English, so the names here are the ones you will see.
OCR: the text in photos and PDFs
- Read on its own: photos are read as they come in, on the phone itself, in every language written in Latin letters.
- OCR again: reads the image again, by blocks: each part of the page in its own paragraph, instead of everything run together.
- Tidying: joins lines broken in the middle of a sentence and words cut at the end of a line, and tidies spaces and stray signs. Line breaks that matter, as in addresses, invoices and lists, stay.
STT: the speech in voice messages, recordings and videos
- One tap: in a record with sound or video, for example a voice message shared from WhatsApp, tap STT, choose the speaker's language or automatic detection, and tap Transcribe.
- No internet: the Whisper engine runs inside the app and knows about 100 languages; the same model serves Portuguese, English, Spanish, French, German and Italian.
- Preferred words: names, places and brands you are going to hear are written beforehand, separated by commas. It is what improves the text the most.
- While it works: a bar shows the percentage done, what is left and when it should finish. On an average phone it takes about 40 per cent of the audio length.
- Reading quality: at the end, how much of the sound became text, silences with no text, repeated phrases and words per minute.
The same pattern in both
- Text tools: a bar below the text shows the words, unknown words and the percentage, and the switches that tidy the text: paragraphs, full stops, lists. They change the text at once and never change what was heard or read.
- Doubtful words: a dictionary for six languages, downloaded once, marks the words it does not know; the Flag button puts them in quotes, which the digital voice does not read, so they are quick to find and fix.
- Use the text: copy, listen aloud, find in search and send by email with the record.
An example
A client sends a four-minute WhatsApp voice message with changes to a quote. The person shares it to Mynd, taps STT and, minutes later, the text is in the record, with the numbers in paragraphs. The next week they find it by searching for a word the client said.
Good to know
- STT does not separate speakers: in a conversation between several people, the text comes out in one piece.
- The dictionary does not catch correct words in the wrong place; that is what preferred words are for.
- The first time, each dictionary language downloads from the Mynd site, a few megabytes; after that it works offline.
The examples show how Mynd can be used; they are not client cases.
