This project leverages OpenSIPS as a SIP gateway, creating a seamless interface between traditional SIP-based communication systems and advanced AI engines. By handling SIP communication and passing voice data to external AI models, OpenSIPS acts as a powerful middleware layer. This setup enables a wide range of voice AI applications, from real-time voice assistants and automated customer support to conversational agents and beyond. While OpenSIPS provides the gateway functionality, it allows developers the flexibility to integrate any AI models needed for tasks like speech recognition, natural language understanding, or voice synthesis. This modularity makes it ideal for building sophisticated, scalable voice-driven applications without being tied to specific AI model constraints.
OpenSIPS functions as a back-to-back SIP endpoint, managing interactions with user agents on one side. On the other side, it connects to an external application — known as the AI Voice Connector — which facilitates communication with the AI engine. This setup allows OpenSIPS to efficiently relay voice data between user agents and the AI engine, ensuring seamless and responsive interactions for voice-enabled applications.
The AI Voice Connector is a modular Python application built to leverage the OpenSIPS SIP stack, efficiently managing SIP calls and handling the media streams within sessions. It provides hooks to capture RTP data, which it sends to the AI engine for processing. Once the AI engine responds, the AI Voice Connector seamlessly injects the processed data back into the call.
Interactions with AI engines can occur directly as Speech-to-Speech if the AI engine provides real-time endpoints. Alternatively, a Speech-to-Text engine can be employed to transcribe the audio. The transcript is then sent to the AI engine as text, and the AI’s response is processed through a Text-to-Speech engine before being relayed back to the SIP user. This flexible workflow allows seamless integration of either real-time voice interactions or a multi-step process that converts speech to text, processes it, and converts responses back into speech for the end user.
The engine is designed to accommodate various AI models, adapting to different AI "flavors" based on each engine's unique capabilities. The currently supported flavors are:
- Deepgram: convert to text using Deepgram Speech-to-Text, push transcribe to OpenAI and then push the response back to Deepgram Text-to-Speech engine
- OpenAI: use OpenAI Real-Time Speech-to-Speech engine
Check out the AI Flavors page for more information.
Engine configuration is done through a separate configuration file, or through environment variables. Using a configuration file is recommended, as it allows for more detailed settings. Also, if you use both methods, configuration file settings will override environment variables. See the Configuration page for all the details.
The simplest way to get the project running is using the Docker Compose files found in the docker/ directory. In order to use them, you need to setup Docker on your host and then run:
git clone https://github.com/OpenSIPS/opensips-ai-voice-connector-ce.git
cd opensips-ai-voice-connector-ce/docker
# edit the .env file and adjust the settings accordingly
# alternatively, create a configuration file
docker compose upAt this point, you should have the engine up and running. A more detailed guide can be found on the Getting Started page.
Then, you can use a softphone like Zoiper or Linphone to send a call to
OpenSIPS by dialling one of the supported flavors (i.e. openai - see flavor
selection). You should be able to talk
to an AI assistent - ask him a question and get a response back.
Documentation pages contain the following topics:
- Getting Started - How to get the engine up and running
- Configuration - Information about configuration file
- Implementation - Implementation details
- AI Flavors - Different AI flavors supported
This project is Community driven, therefore any contribution is welcome. Feel free to open a pull request for any fix/feature you find useful. You can find technical information about the project on the Implementation page.
The OpenSIPS AI Voice Connector Community Edition source code is licensed
under the GNU General Public License v3.0
All documentation files (i.e. .md extension) are licensed under the Creative Common License 4.0
© 2024 - OpenSIPS Solutions
