Cabinet
console.gapi.uz is where projects are created, options set, tokens issued and every recording seen. Sign in with Google.
Language
The cabinet speaks Uzbek, Russian and English, and opens in Uzbek unless it is told otherwise — arriving from gapi.uz/en or gapi.uz/ru opens it in that language. The globe in the top bar switches it at any time; the choice is remembered in the browser, and the documentation opens in the same language. The names quoted below are the ones the English interface shows.
Project
One account can hold several projects: per customer, per integration, per department. Each has its own settings, tokens, voices and recordings.
Project tabs
The tabs are listed in the order the cabinet shows them.
Recognition. Drop a recording and recognition starts at once: first the "Upload" step with the percentage of the file, then "Recognition" with the seconds of audio. The result opens on the same page: text by utterance, speakers, a player, a "hide personal data" switch. The "Transcription options" block changes settings for the recordings you send next; the same block asks whether it is a call and its direction. At the bottom, "Connect over the API": a project with no token gets one made for it, and the examples carry your own token.
Recordings. Every job of the project: state, length, time. Clicking opens the result. The list comes fifty to a page, newest first.
Vocabulary. The project's terms: company and product names, people's names. The model starts recognising them in speech more often. One term is enough and twenty is the limit: the shorter the list, the more often the model lands on the word you meant. A vocabulary version is activated with one button. In detail: Vocabulary.
Speakers. On the left, the project's voices in three groups: "Waiting for a name" — heard in three recordings or more, "Named" — kept for good, and "Candidates" — heard once, gone in 30 days. On the right, the panel for the selected voice: listen, name it, and see where it was heard — every recording it spoke in, with how long it spoke; a deleted recording is still listed, it just cannot be played. Two doors at the top: "Record" opens a dialog with a sentence to read aloud, the microphone and a half-minute cap; "Upload a sample" takes a file or a recording the project has already transcribed. Both ask the same thing — whose voice this is. A name the project already knows adds a print to that person; a new one starts a voice. In detail: Voices and speakers.
API tokens. API keys. A key sits covered in the table: "Show" uncovers it, and "Copy" copies it whole even while it is covered. You can see when each was last used.
Settings. Every option of the project in two groups: "Who is speaking" and "Paid analysis". The price sits next to each option: only Emotion and Personal data multiply the seconds, the rest are free. The text options — punctuation, numbers, stereo channels — are always on and are not shown here. Each option has "how this works": what it does, what the recording needs, where the answer is in the JSON, the price. The tariff is computed live. The maximum number of speakers and the webhook address are here too.
The player
A result opens as a timeline, the kind an editing suite has: time runs left to right, one lane per speaker, every utterance a block. The colour is the mood, the yellow dot is personal data. Click a block and it plays that utterance in full, then moves on to the next by itself. The transcript below never scrolls on its own: the page stays where you left it.
| Key | What it does |
|---|---|
space, k | play or pause |
← → | five seconds back or forward |
Shift + ← → | previous or next utterance |
+ − | stretch or shrink the timeline |
0 | fit the whole recording |
Ctrl + wheel | zoom around the pointer |
| wheel | move the timeline sideways once it is stretched |
While the timeline is stretched it follows playback; the "follow" button turns that off.
Usage and limit
The balance in g shows under your name in the left menu and on the "Billing" page with its history. During the beta it is topped up to 240 g every midnight Tashkent time. Paid options spend it faster: Emotions twice as fast, Personal data by 20 %. In detail: Balance and billing.