What the Studio is

The Studio is Uttera's web application: a page where you upload a file or paste some text, tick what you want and download the result. Nothing to install, no key to request, and not a line of code to write.

Underneath it is exactly the same service the rest of this documentation describes. The Studio is not a demo with trimmed limits: it calls the same API, on your same plan, and spends credits from your same allowance. When this documentation says transcribing costs so much per second, it costs the same through both doors.

To get in you need an account and an active session. If you don't have one, the Studio sends you back to the home page. The page header reminds you which email you signed in with and which plan you're on, because what you can do depends on it.

How the screen is laid out

There are four cards, one per job:

If your plan doesn't include something, the card is still there but switched off, greyed out and without a button, saying which plan you'd need. It's shown on purpose: it's better to see what exists and what it costs than to discover it through an error. The same happens inside Transcribe with the analysis and summary checkboxes, which switch off individually without taking the transcription away.

All the cards work the same way, top to bottom:

1. The fields. The file or the text, and that job's options.
2. The cost line. It appears as soon as you choose a file and says what it's going to cost, before spending anything.
3. The Run button.
4. The result. While it works you'll see the Uttera logo animated and a seconds counter, so you can tell it's still alive; long audio takes time. When it finishes, the logo goes still and the counter turns into how long it took.

Transcribing audio

The most complete card: it transcribes, and along the way it can analyze and summarize.

1. Choose the file. It accepts wav, mp3, flac, ogg, opus, aiff, m4a and webm. Up to 400 MB and 3 hours of audio.
2. Tick the extras you want (each has its price and is billed separately):

   Checkbox	What it adds
   tone	The dominant emotion of the recording, with its confidence and the other candidates.
   speaker profile	Age range and gender estimated from the voice.
   who speaks and when	Speaking turns with timestamps per speaker.
   summary of the recording	A structured summary of what was said.

3. Run. The file is uploaded only once even if you tick all four: the analysis is distributed internally over the audio already uploaded.

The result is always laid out in this order:

1. The transcript, and below it the tone as a line of prose if you asked for it, with the Download text button (transcripcion.txt) closing the block.
2. The summary, if you asked for it, with Download summary (resumen.txt). It goes after the text on purpose: a summary is easier to follow with the transcript in front of it, and the transcript is what's always there.
3. The profile and who speaks and when, raw. They're structured data and whoever asks for them wants them whole, not written up.
4. Download analysis (analisis.json): a single file with everything that was analyzed — including the full tone vector — not one per section.
5. The charge line, which closes the operation.

If a particular analysis fails, it is said: the transcript is delivered anyway and below it appears which part couldn't be done. It isn't hidden and it isn't made up.

Text to digital voice

Write or paste a text and choose how you want to hear it.

The result is a player, the charge line and Download audio (voz.mp3, or whichever extension you chose).

Text to cloned voice

Same as the previous one, but you supply the voice: you upload a sample and the text is read in it.

The sample is not charged. What's charged is the generated audio, which depends on how long the text is, not on how long the sample is. Cloning takes considerably longer than the digital voice: the card warns you while it works. The download comes out as voz-clonada.mp3.

Translating a recording

The whole chain in one step: it transcribes the audio, translates the text and says it again in the target language.

The result shows the transcribed original and the translation, and below them the player with Download audio (traduccion.mp3) if you asked for audio. If any part of the chain had something to warn about, it appears as a notice instead of getting lost.

What it costs, before and after

The Studio is also where you learn what things are worth, so it states the price twice: an estimate before and the real charge after.

Before. As soon as you choose a file, the browser reads its duration and multiplies it by the price of what you've ticked. Ticking or unticking a box recalculates it without having to choose the file again. It says approximate cost because it is: the duration a browser measures on a variable-bitrate mp3 isn't always exact, and it's the engine that bills.

The two voice cards carry no advance estimate, and that isn't an oversight: they are billed by seconds of audio generated, and how much audio a text will produce isn't known until it's generated.

After. When it finishes, a line appears with the time it took to process and what was consumed; a second later it is replaced by the real charge, exactly as the billing system recorded it:

29 s · Charged: 144 credits · for 2,424.0 s of audio ·
transcription 79 · tone 12 · profile 8 · speakers 24 · summary 21

The prices this screen uses are the same coefficients published in this documentation, read live: if they change, they change here at the same time.

When something goes wrong

The Studio's messages are the API's errors in plain language:

And if what you want is for all of this to happen on its own, with nobody uploading files by hand, that's the other door: Integrating Uttera into your code.