We build speech technology

SOTA AI is a team from Tashkent. Our own speech recognition and synthesis models, and the products built on them for banking, telecom and the public sector.

Write to us

We teach machines to understand the language people actually speak: Uzbek, Russian, and whatever comes out when the two mix in a single sentence.

Why we do this

The world’s speech systems are trained on English, Chinese and Spanish. There is almost no Uzbek in their data, and no mixed speech at all — the kind where a sentence starts in Uzbek and ends in Russian. An off-the-shelf model stumbles over such a sentence where a person understands it instantly.

We did not wait for the big companies to get around to our language and built our own models instead: recognition, synthesis, biometrics, meaning. We collected and labelled the data ourselves, from real conversations, with the consent of everyone involved.

Every one of our products runs on these models: Speech Analytics, the AI Prompter, the AI Trainer, Voice Biometrics, bots, the AI Agent Factory, Meeting Minutes and Audio Badges.

Three directions

Speech technology

Speech recognition and synthesis in Uzbek and Russian, including mixed sentences. Speaker separation, punctuation, timestamps and a streaming mode for telephony.

Learn more

Voice Biometrics

Identity verification by voice without passphrases: the customer simply talks. For banks, insurers and public services.

Learn more

AI agents

Bots, assistants, the Prompter and the Trainer. All built on one platform that understands both languages and installs inside your own network.

Learn more

How the platform is built

We develop the speech models ourselves, from recognition and synthesis to understanding meaning. All eight products run on them, so they see the same conversations and speak the customer’s language.

5speech synthesis languages, a new one on request

50+languages for text processing

95 %speech recognition accuracy

Many languages

Speech synthesis in five languages, with a new language developed on request. Text processing in more than 50. We understand Uzbek and Russian even when they are mixed in one sentence, and we also recognise Kazakh and English.

Accuracy and speed

Speech recognition accuracy of 95 %. Real-time mode over gRPC and MRCP: the transcript appears during the conversation, not after it.

Every channel at once

The AI reviews 100 % of communication — calls, chats, video meetings, face-to-face conversations and service enquiries — by one and the same set of rules.

Flexible deployment

On-premise, with licences for the server and for seats, or cloud on subscription. In the first case the recordings never leave your network.

Straightforward integration

The speech technology plugs into your own applications and services: telephony, CRM, messengers, internal systems. No separate workplace is needed for it.

Reliability

Continuous platform operation and data stored in a TIER III data centre. Access is split by role and personal data is masked in transcripts.

Who we work with

Companies that talk to their customers a lot, and in two languages.

The company in numbers

20 yearsin speech technology

300+rollout projects

95 %speech recognition accuracy

Office

Our office is on Mustaqillik Avenue in Tashkent. Come and visit: we will show the products on live data and introduce you to the team.

Legal entity: ООО «SOTA-AI»
TIN: 310642723
Address: Tashkent, 7 Mustaqillik Avenue, Oqibat neighbourhood, Mirzo-Ulugbek district
Status: Resident of IT Park Uzbekistan

All contacts →