Problems in french
# support
c
I am trying to use VAPI to create a voice assistant but I am encountering several issues: I am looking to create a voice assistant in French: I have significant costs related to translation with TalScriber. Are there any alternatives that might be less expensive? DeepGram seems to be having difficulties, or at least it doesn't seem to work as well. Maybe I am using it incorrectly. The biggest problem is latency; between the ElevenLabs voice and the translator, I am at 2200 ms, which seems unusable. Do you have any recommendations to improve this? If not, are there any upcoming updates to VAPI that will address these issues? Thank you in advance!
v
Hi @Celemiba, I'm an AI assistant for Vapi. I'll analyze available sources and get back to you shortly.
----------------------------- **While I generate an answer, here are some sources that might be helpful**: * [Estimating Costs]() * [Web Calling]() * [Billing Examples]() * [Outbound Calling]() -----------------------------
### Answer Creating a voice assistant in French with Vapi can indeed present some challenges. Here are a few suggestions and insights to help improve your setup and address the issues you're encountering: ### Alternatives to TalScriber for Cost Efficiency: 1. **Deepgram**: Although you mentioned difficulties, Deepgram is generally known for lower costs. Ensure you're using the appropriate model and language settings. Unfortunately, Deepgram does not currently support Arabic, as noted in a forum post, but it does support French and could be a good option if configured correctly [(1)]() [(2)]() . ### Reducing Latency: 1. **Voice Provider**: ElevenLabs is known for high-quality voice synthesis, but it can add latency. You might consider testing other providers like PlayHT or Azure, which may offer faster response times. PlayHT is often used for its conversational voices [(3)]() [(4)]() [(5)]() . 2. **Optimizing Configuration**: Ensure your configuration is optimized for minimal latency. For example, using models and providers that are geographically closer to your target audience can help. ### Recommendations: To reduce overall latency and costs, you could try experimenting with different configurations and providers. For example, using Deepgram for transcription and PlayHT for TTS might offer a balance between cost and performance. ### Future Updates: While I don’t have specific information on upcoming updates, Vapi is continuously improving its services to address common challenges, including latency and cost efficiency. Stay tuned to their announcements for updates on improvements
and new features [(6)]() [(7)]() [(8)]() . ### Example Configuration: Here's an example of setting up an assistant using Deepgram and PlayHT:
Copy code
javascript
import Vapi from "@vapi-ai/web";

const vapi = new Vapi("your-public-key");

const assistantOptions = {
    name: "French Assistant",
    firstMessage: "Bonjour, comment puis-je vous aider?",
    transcriber: {
        provider: "deepgram",
        model: "nova-2",
        language: "fr",
    },
    voice: {
        provider: "playht",
        voiceId: "jennifer",
    },
    model: {
        provider: "openai",
        model: "gpt-4",
        systemPrompt: "You are a French-speaking assistant...",
    },
};

vapi.start(assistantOptions);
[(3)]() [(9)]() [(4)]() [(5)]()
s
Which LLM model you are using?
c
Groq + llama 8b to reduce costs
s
You can try azure voice provider
c
Already tried however the quality is very low in French
Robotic voice
s
Man, then you need to compromise with the transcriber, or you can wait for a couple of days until OpenAI launches voice-to-voice.
c
That’s exactly the reason why i be asked if any updates were coming about this ! Ty
9 Views