What gapi is
gapi turns recordings of Uzbek conversations into text and tells you what happened in them: who spoke, about what, in what mood, and which personal data was said aloud.
How it works:
- You upload a recording, through the cabinet or the API.
- A few seconds later the result is ready: text by utterance with timing, speakers and the options you switched on.
- You read it in the cabinet, fetch it by link, or receive it through a webhook.
What it can do
| Capability | What you get | Tariff |
|---|---|---|
| Speech recognition | Text with the time of every utterance. Any recording in Uzbek: a call, a meeting, an interview, a dictaphone; in Latin script | ×1 |
| Term vocabulary | Company and product names and people's names are recognised more often | ×1 |
| Punctuation, numbers | Text that reads as written: punctuation, capitals, digits | ×1 |
| Who is speaking | The sides by stereo channel, or voices separated in mono | ×1 |
| Voice gender, client and manager | Which party is which | ×1 |
| Recognise by voice | The same person gets the same number in every recording | ×1 |
| Emotions | The mood of every utterance and a summary per speaker | ×2 |
| Personal data | Names, phones, addresses, documents found and masked | ×1.2 |
The tariff multiplies the length of the recording; it is not a charge per option. Every ×1 row can be on at once and an hour of audio is still an hour. Only Emotions and Personal data make it cost more.
Every option in detail: Options.
Where to start
- Quick start: a token, the first recording, a result in five minutes.
- Use cases: ready-made option sets for calls, meetings, interviews.
- Result format: what the JSON holds.
- API: every request, limits, webhooks.
What it costs
The unit is g: 1 g is a minute of audio at ×1. An hour of recording without paid options costs 60 g. Emotions double the bill, Personal data adds 20 %, and together they make ×2.2. During the beta every account's balance is topped up to 240 g, four hours of audio, every midnight. In detail: Balance and billing.