Voice transcription
BetaDictate instead of typing — audio in, punctuated text out, English or Hindi. It transcribes; it does not answer.
- ✓Audio-to-text with automatic punctuation and formatting
- ✓English and Hindi
- ✓25 MB per file
- ✓Transcription sessions listed in your history; the audio itself is not kept, only the transcript
What it is.
Send an audio recording and get back a formatted, punctuated transcript, in English or Hindi. Useful for dictating a question, a note on a contract, or a file note without typing it. Transcription sessions are kept so you can go back to what you dictated.
This is speech to text and nothing more. It does not send the transcript to the AI, it does not answer your question, and there is no synthesised speech coming back — there is no text-to-speech anywhere in the product. Paste the transcript into Vio to get an answer. It is also not Meeting Intelligence: this is a dictation of one voice into a text box, while a meeting is a record — chunked live transcription, inferred speaker labels, minutes, tasks, an export, a place on the matter timeline.
Three steps.
End to end.
Upload an audio file, up to 25 MB.
English or Hindi.
Punctuated and formatted text, with the session kept in your history.
What you get.
- ✓Audio-to-text with automatic punctuation and formatting
- ✓English and Hindi
- ✓25 MB per file
- ✓Transcription sessions listed in your history; the audio itself is not kept, only the transcript
- ✓A microphone in the phone's text boxes, and in New Contract, where speech becomes a drafting brief
- ✓Disabled outright in air-gapped deployments, like every other outbound vendor
Quick answers.
No. It returns text. There is no speech synthesis in the product and no automatic hand-off from a transcript to the assistant — paste the transcript into Vio yourself.
The language options are English and Hindi. There is no separate Indian-English acoustic model, so heavily accented English will transcribe as well as a general English model manages, and no better.
Yes, as sessions in your account, under the same retention and deletion terms as your other content.
No — use Meeting Intelligence. Voice transcription takes one file of up to 25 MB and gives you text. A meeting is recorded in chunks as it happens or uploaded as a file of up to 300 MB, gets speaker labels (inferred, never identified), minutes, action items that become tasks, a PDF or Word export, and a place on the matter's timeline. None of that exists here.
More in Platform & Deployment.
Storage and processing default to AWS ap-south-1 (Mumbai). Model inference leaves India — that is disclosed, not hidden.
Tax invoices numbered consecutively within the financial year, with the CGST/SGST or IGST split, or zero-rated under LUT.
A read-only Indian legal-data API — clause benchmarks, judgment search, statute search — metered per key.
Right-click any PDF or Gmail attachment → "Scan with LexVio".