PatientDesk launches Turkish voice models Duyu and Alania-1

PatientDesk, which builds an AI front desk for dental clinics, has released two voice models for Turkish. Duyu transcribes spoken Turkish, and Alania-1 turns written Turkish into speech. Both work with the same API key from a single account. Co-founder Oncel Ozgul said in the announcement that the team wants to support work on Turkish-language AI.

What Duyu is and what it does

Duyu is a speech-to-text model for Turkish. It can transcribe live audio as people speak, and it can also process recorded files. During a conversation it sends interim text, then the final version when each segment ends. It can optionally return word-level timestamps and subtitle files in SRT or VTT format.

The company's site highlights call-center situations such as rescheduling appointments, reading out addresses and spelling names letter by letter. You can try Duyu on the site by speaking into your own microphone.

What Alania-1 is and what it does

Alania-1 is a text-to-speech model for Turkish. According to the company, it was trained on Turkish speech data and works out sentence stress and pauses from the text itself, with no extra markup needed. Its highlighted features include reading dates, times and numbers clearly and changing tone for questions. Audio can be streamed while it is being generated, so an app can start playing the reply right away. Alania-1 uses a single voice and marks every response as AI-generated.

The company's claims

According to the company, Duyu scores highest in Turkish among the open-source and leading models it compared on FLEURS and conversational benchmarks. The site shows tables comparing Duyu with OpenAI's Whisper models and Alibaba's Qwen3-ASR models on three Turkish datasets made up of conversation, read speech and call-center recordings. The company also says both models respond fast enough for phone calls.

How to use it

Both models are used through PatientDesk's voice API. Duyu's file transcription endpoint is compatible with OpenAI's audio API, so apps that already use that format can connect with a small change. Live audio runs over WebSocket. Signing up does not require a card, and as a launch offer both models are free for one month, with pay-as-you-go pricing after that. For enterprises, the company also offers deployment on the customer's own infrastructure. Current pricing is on PatientDesk's site.

PatientDesk was founded in 2025 by three Turkish founders, Oncel, San and Emre, and is part of Y Combinator's Winter 2026 batch. Its main product is an AI front desk that answers dental clinics' phones and books appointments, and Duyu and Alania-1 grew out of that work. The comparisons shared are the company's own measurements. To see how the models handle a wider range of Turkish recordings, it is worth following evaluations from independent researchers as well.

Sources: PatientDesk Turkish voice models, Duyu, Alania, Y Combinator: PatientDesk.