Using Moshi (Voice to Voice)

To test Moshi, you need to set up and run the Moshi Server on Runpod (Or your running it on your own computer/server):

---

Step 1: Set Up Moshi Server

1. Login to the Terminal on your instance, whether it is your own server or a Runpod instance (If you don't have a good GPU)

2. Clone the Moshi Server

   git clone https://github.com/flukexp/moshi_server.git moshi && cd moshi

3. Create a virtual environment

   python -m venv venv

4. Activate the virtual environment

   • On macOS/Linux:

     source venv/bin/activate

   • On Windows (Command Prompt):

     venv\Scripts\activate

   • On Windows (PowerShell):

     .\venv\Scripts\Activate

5. Install dependencies

pip install -r requirements.txt

---

Step 2: Start the Moshi Server

Start the Moshi server (For Runpod):

uvicorn moshi_service:app --host 0.0.0.0 --port 8000 

Start the Moshi server (For your own computer):

uvicorn moshi_service:app --host 127.0.0.1 --port 8000 

---

Step 3: Change Settings on Amica to Use Moshi

Open Settings > Chatbot Backend , and select Moshi.

Then go to Settings > Chatbot Backend > Moshi, and then insert the correct URL for accessing Moshi server. (E.g. http://localhost:8000 for local server and a runpod proxy URL, which is on your runpod instance and looks like : https://rn8xojhvb-8000.proxy.runpod.net, the URL has the runpod instance identifier and the port)

There is no difference whether you are running locally or off the Amica demo.

---

Notes:

• Ensure that Python and pip are installed before proceeding. Edit the model URLs in the main python script if you want to use a different model from Kyutai.

---

Replace the {port-number} in the URL with the actual port number.

---