OpenAI has released GPT-Live-1.1, a new version of its voice models, adding 13 new voices and enhanced speech in a handful of non-English languages. According to the company, the improvements include better speech performance in Hindi, French, German, Korean, Japanese, Mexican, Spanish, and other languages. OpenAI announced the voice models update on October 15, adding that the update also makes delegation more natural and improves user steering when co-piloting with the model. The update is being rolled out to ChatGPT users, with Codex and dots also announced. API access is on its way, but already some users are experiencing issues with voice selection and slight pauses.
Improved Hindi, French, German, Korean, and Japanese Support
Performance is the most significant update in GPT-Live-1.1 for non-English languages. Notable examples OpenAI mentions are Hindi, French, German, Korean, Japanese, and Mexican Spanish. There are also new English voice choices in the release.
Those who speak Hindi may find voice interactions easier to use when they want to ask questions, exchange ideas, or perform common tasks. Other speakers of French, German, Korean, and Japanese can also benefit from similar improvements when using the model without having to depend on English. Accent quality is also part of the update. Even if a voice assistant understands your words, it could still have a poor accent when speaking a new language. Better pronunciation and speech quality can make listening to spoken answers more comfortable.
What the New Voices Mean for ChatGPT and Codex
The new voices provide users with more options for selecting the voice of ChatGPT when chatting with the language model. The release notes from OpenAI list that more voice choices may be available, such as for various languages, in addition to English.
In Settings, select Voice and tap a default voice to see what options are available to you. The update is gradually being rolled out, so not all users can have the same options across all of their accounts at the same time. Voice interaction, which allows you to give instructions as you complete coding tasks, is among the announced support. However, there is no separate performance breakdown for Codex. Dots, which is OpenAI’s product for personal agents, is also part of the update.
Voice Glitches Surface During the Launch
Not all users have been able to have a seamless experience. Days after the announcement, some users in the OpenAI Developer Community noticed issues with the voice choice in Standard and Advanced Voice mode on Android and iOS. Some say they are hearing a voice that does not match the one chosen in the settings; others say they’re transitioning between the two voice modes unexpectedly.
Other issues involve multiple pauses and inaccurate language recognition. For instance, in one case German speech is sometimes being detected as an odd assortment of other languages, while in another case a previously selected voice has stopped working. These accounts describe individual user experiences and do not confirm the extent of the problems.
When Will GPT-Live-1.1 Reach API Users?
OpenAI states that the rollout to users is currently in progress and the API will be available shortly. So, in other words, don’t assume you have the new model through the API just because you see new voices when chatting in ChatGPT. API version might also offer another option for developers to design voice applications with the new model, although they’ll have to wait to see when it’ll be released and how that will impact an integration.
For now, GPT-Live-1.1 offers additional voice options, enhancement in multilingual speech, and greater talk control. The release also demonstrates the fact that advancing voice AI is more than just supporting fresh voices. Consistent voice selection, accurate language detection, and fewer normal disruptions are other benefits.
