Transcription
How a spoken recording becomes accurate, readable text without losing what was said.
After a guest records, Dafado turns the audio into text. This is what makes a review searchable, translatable, and readable at a glance, while the voice itself remains the thing you can play back and trust.
Speech to text, in the language spoken
Recordings are transcribed by a speech-to-text engine that listens to the audio and writes down what was said. It detects the spoken language automatically, across many languages, so a guest never has to declare what language they are about to speak. Someone talks, and the transcript comes back in the language they used.
The transcript is kept in the language it was spoken. Reading it in another language is handled separately, by translation, and always starts from this original text.
A cleanup pass that stays conservative
A speech-to-text engine hears sound, not meaning. It has no knowledge of the world, so it can mishear an unusual name or a specific term. To catch this, Dafado runs a cleanup pass after transcription.
The cleanup is deliberately careful. It proposes only small, specific corrections, such as a garbled word that is clearly a misheard name, or the capitalization of a name. Each proposed correction is then validated before anything is applied, so a suggested change that would alter meaning is rejected rather than trusted. The aim is to fix obvious mishearings while preserving exactly what the guest meant.
Some mishearings are left for a person to fix. When the engine mishears a name as an ordinary, valid word, the careful cleanup will not touch it, because changing valid words by guesswork would do more harm than good. Those cases are corrected by hand instead.
The raw recording is never altered
One rule sits underneath all of this: the raw recording is never changed. It is the original the transcriber reads, and it is kept as it was made. Corrections apply to the text, not to the audio.
This keeps the voice honest. Anyone can play back the actual recording and hear the words for themselves, and because translations are generated from the corrected transcript, fixing a mishearing at the source improves every language at once.