AI Voice Agent Self Host

We need a Turkish-speaking AI phone agent. The telephony is already ours we have our own call API and we handle the calls. What we need is the conversation layer and the agent side: speech in, speech out, and a way to build, train and manage the agents themselves. Three constraints: 1. Turkish must sound natural on a phone call. 2. Self-hosted / private no ElevenLabs or similar third-party voice APIs. The model runs on our servers, commercial-use license. 3. We must control pronunciation— record a short sample of how a word should be said (brand names, amounts, dates) and have the agent say it that way. We're also looking for guidance, not just code: which product/model we should use, how to train and tune an agent, what to do as we grow. We may move to a semi-custom setup later, so we want someone who can stay with us. We can do the development part with dedicated team. We need someone to consult and be able to take actions on the project when its stucked. Tell us what you can offer.