Bug / Unexpected behavior: Marin / Cedar voices repeat greeting & no dialect (Flemish)

API Bugs
1

Bug / Unexpected behavior: Marin / Cedar voices repeat greeting & no dialect (Flemish)

We’re using the new realtime voices in production (Marin and Cedar) via the Realtime API / voice agent. Since switching, we’re seeing two unexpected issues:

1. No dialect support (Flemish / Dutch)

In our region (Dutch Belgium) we speak Flemish. Previously we used the Coral voice and it could handle Dutch/Flemish quite well.

After switching to Marin or Cedar, the voice only speaks standard Dutch. No Flemish anymore, which sounds less “local” to our users.

2. Greeting is spoken twice

When a call starts, we send a single conversation.item.create with the greeting text. This works correctly for all voices:

const greetText =
  session.greetingMessage ||
  `Greet the caller in their own language, using a time-of-day awareness. For example, if it is morning in their local time zone, say "Good morning," if it is afternoon, say "Good afternoon," and so on (even if the language is not English, adapt accordingly). If you know the caller's name from the EndUser information, address them by name, e.g. "Good afternoon, Thierry De Decker, welcome to iPower. I am AnswerPal, your 24 on 7 digital assistant. How can I help you today?" If no name is available, greet them in a polite, friendly manner. Mention that this conversation is being recorded for quality assurance purposes. Never repeat the greeting.`;
const greetEvent = {
  type: "conversation.item.create",
  item: {
    type: "message",
    role: "user",
    content: [
      {
        type: "input_text",
        text: greetText,
      },
    ],
  },
};

With Marin or Cedar, as soon as the caller says something, they first repeat that exact greeting and only then provide the “real” AI response. The caller therefore hears the greeting twice: once when the agent picks up, and again right before the first answer.

It feels like Marin/Cedar interpret the greeting instruction differently.

3. Questions for the community

  • Do Marin or Cedar speak a dialect for you (e.g. Flemish or other regional variants)?

  • Do you also experience double greetings or other prompt repetitions with these voices?

  • Are there known workarounds or best practices (extra instructions, specific session config, “skip greeting on first user utterance”, …), or does this look like a bug in the voice engine / session flow?

1 Like