Is there a setting to make my assistant speak slower?

Hi there! 👋 Yes, there are definitely settings to make your assistant speak slower in Vapi! Let me explain your options. Most voice providers in Vapi offer speech speed controls through the

speed

parameter. Here's how you can adjust it depending on your voice provider: - **OpenAI voice provider**: Supports speed settings up to 4x, which means you can set lower values for slower speech

const voiceConfig = {
    provider: 'openai',
    voiceId: 'alloy',
    speed: 0.8,  // Lower values = slower speech
  };

[Source](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1339223223409184788) - **RimeAI voice provider**: Has flexible speed settings with no upper limit

const voiceConfig = {
    provider: 'rime-ai',
    voiceId: 'abbie',
    speed: 0.7,  // Lower values = slower speech
  };

[Source](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1339223223409184788) - **Cartesia voices**: You can specify voice speed using

Assistant.voice[provider="cartesia"].experimentalControls

[Source](https://docs.vapi.ai/changelog/2025/2/25) You can access these settings through the Vapi API, as "speech configurations are currently only made via API" according to the [Speech Configuration documentation](https://docs.vapi.ai/customization/speech-configuration). Please note there are some limitations with certain providers: - When using the OpenAI Realtime API (4o or 4o-mini), the speed settings might not be effective [Source](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1339276020389253181) - For Eleven Labs, there is no direct way to control the speed of voice [Source](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336023375721791549) Is there a specific voice provider you're using that I can provide more detailed instructions for? Sources:

I just have to paste that json? @Vapi

If your voice provider has the feature to control the voice speed, then you can definitely do it through SSML parsing. If not, Sorry, then there is no option for it.