OpenAI has finally released its much-anticipated 'Advanced Voice Mode' (AVM) for ChatGPT, after facing several delays reportedly linked to safety concerns and optimization. This feature, which was first announced and demonstrated in May, allows users to engage in real-time conversations with the ChatGPT artificial intelligence model using a text-to-speech synthesization module.
Historical Context: Duplex’s Influence
Those acquainted with the technology landscape may recall Google's Duplex AI service, unveiled in 2018 with promises of imminent availability. The service showcased an AI system capable of autonomously calling businesses to schedule appointments on the user's behalf, interacting with humans in real-time. While the Duplex project was eventually discontinued, its legacy appears to have influenced OpenAI's ChatGPT development.
Exploring the Advanced Voice Mode
The AVM feature introduced by OpenAI offers real-time communication designed to replicate human-to-human dialogues. ChatGPT provides responses in a human-like voice with a natural flow, allowing users to interrupt the chatbot mid-conversation. Moreover, according to demonstrations, the system can remember previous interactions, enhancing the conversational experience.
Safety and Rollout
OpenAI emphasizes safety as a paramount concern in the deployment of AVM. The company conducted extensive testing with over 100 external 'red teamers' across 45 languages to ensure user privacy and prevent misuse. The model is programmed to adhere to specific preset voices, with mechanisms in place to block inappropriate content requests.
Image Source: OpenAI
In a statement on the feature release, OpenAI stated:
'We tested GPT-4o's voice capabilities with 100+ external red teamers across 45 languages. To protect people's privacy, we've trained the model to only speak in the four preset voices, and we built systems to block outputs that differ from those voices. We've also implemented guardrails to block requests for violent or copyrighted content.'
The rollout of AVM has commenced, with additional users to be gradually added in the future. OpenAI plans to make this feature available to all Plus subscribers by fall.















