OpenAI Offers Clue to Voice Assistant Plans

Recent trademark filings suggest that OpenAI, the maker of ChatGPT, could be in the process of developing a voice assistant product. The trademark application, filed on March 19, covers the term "voice engine."

The application is for a service mark, indicating its association with services rather than tangible goods, which could suggest a new product is in the works.

While companies often file trademarks that never translate to real products or services, OpenAI already offers voice capabilities baked into ChatGPT through its Whisper speech recognition system.

ChatGPT’s voice functionality, which is available for free from the app, allows users to have back-and-forth conversations with ChatGPT.

Companies can also access Whisper through the company’s API to build commercial applications, like for transcribing customer conversations in call centers.

Despite the filing, OpenAI has not confirmed it’s working on a voice assistant offering akin to Apple’s Siri. The Microsoft-backed company is looking beyond text generation to multimodality through new underlying models like Sora for video generation.

One project OpenAI is working on its GPT-5, its next-generation foundation model. While no official time frame has been published on when it will launch, there is widespread speculation that GPT-5 will be revealed later this year. GPT-4 came out last March while GPT-4 Turbo was unveiled during the company’s DevDay in November.

Related:OpenAI’s Whisper v3: Improved Speech Recognition for Business Applications

Having already secured a trademark for its next model, OpenAI also filed applications for GPT-6 and GPT-7 in October 2023. These are also service marks covering software including “computer software for using language models.”

The trademarks for GPT-6 and GPT-7 provide insights into the potential capabilities of OpenAI's future language models.

OpenAI's trademark filings for GPT-6 and GPT-7 list goods and services that include multilingual speech recognition, enabling the transcription of speech across multiple languages, as well as generating music – abilities current GPT models do not possess.