In a groundbreaking collaboration, ElevenLabs successfully integrated its advanced conversational AI platform with DeepSeek’s R1 model. Now, instead of typing text prompts with the popular Chinese chatbot, you can naturally chat with the agent’s human-like voice.
This fusion not only enhances user interaction but also showcases the versatility of ElevenLabs' technology in adapting to various large language models (LLMs).
ElevenLabs' conversational AI platform
ElevenLabs has developed a state-of-the-art Conversational AI platform designed to deploy customized, real-time conversational voice agents. Flexibility is a key feature that allows seamless integration with various LLMs through OpenAI-compatible APIs. This adaptability ensures it can cater to diverse applications, from customer service bots to educational tutors.
Since DeepSeek recently unveiled its R1 model, it has become a prominent player in the AI industry. Renowned for its exceptional reasoning capabilities, particularly in complex fields like mathematics and coding, the R1 model employs a "chain-of-thought" reasoning approach, enabling it to process and solve intricate problems step by step. Yet, it lacked a natural language interface, which limited accessibility to users.
The integration process
To voice-enable DeepSeek R1, ElevenLabs utilized the DeepSeek-R1-Distill-Qwen-32B model, a distilled version of R1 based on Qwen2.5. This model was chosen due to its support for function calling, a crucial feature for interactive conversational agents. The integration process involved several steps, including agent creation. Within the ElevenLabs platform, “DeepSeeker,” a new AI agent, was born using the Math Tutor template. This template was selected mainly because it highlights the reasoning capabilities in the mathematical contexts of DeepSeek.
ElevenLabs also leveraged Cloudflare’s Workers AI platform. The DeepSeek-R1-Distill-Qwen-32B model was accessed via an OpenAI-compatible API. The necessary API keys and endpoints were configured within the ElevenLabs platform to establish a seamless connection. The system prompt was customized to ensure that the AI would articulate numbers and equations in natural language, enhancing clarity for users.
Applications and future prospects
This integration results in a voice-enabled AI agent capable of engaging users in natural, spoken dialogue. I tried it myself and had a lot of fun asking it questions, hearing it “think” aloud, and responding to my questions. Even a simple “How are you?” prompted it to curate a response to keep our conversation going carefully.
But beyond general conversation, the chatbot holds potential across various sectors. For instance, in a mathematical tutoring scenario, the agent can guide students through complex problems, articulating each step clearly and concisely.
This successful integration underscores the potential of combining advanced reasoning models with natural language interfaces. As AI continues to evolve, such collaborations pave the way for more intuitive and accessible AI applications, bridging the gap between complex computational models and everyday user interactions.
