In the AI market of overcrowded voice, OpenAi is betting on monitoring of instructions and expressive speech to gain business adoption
Do you want smarter information in your reception box? Sign up for our weekly newsletters to obtain only what matters for business managers, data and security managers. Subscribe now
OPENAI adds to an increasingly competitive AI vocal market for companies with its new model, GPT-Realtime, which follows complex instructions and with voices “which seem more natural and expressive”.
While the voice of AI continues to grow and customers find use cases such as customer service calls or real -time translation, the AI votes market with realistic consonance which also offer business quality security is warmed up. Openai says that her new model provides a more human voice, but she still has to compete with companies like Elevenlabs.
The model will be available on the API in real time, which the company has also made available. With the GPT-Realtime model, Openai also published new voices on the API, which he calls Cedar and Marin, and updated his other voices to work with the latest model.
OPENAI declared in a livestream that he worked with his customers who build vocal applications to form GPT-Realtime and “alignment carefully the model with estuals that are built on real world scenarios such as customer support and academic tutoring”.
The AI scale reached its limits
Electricity ceilings, increase in token costs and inference delays restart the AI company. Join our exclusive fair to discover how best the teams are:
- Transform energy into a strategic advantage
- Effective inference architecting for real debit gains
- Unlock a competitive return on investment with sustainable AI systems
Secure your place to stay in advance::
The company has boasted the ability of the model to create emotional and natural consonance voices that also align with the way developers build with technology.
Speech speech models
The model works in a speech framework, allowing it to understand the guest prompts and to respond vocally. Speech -to -speech models are perfectly suited to real -time responses, where a person, generally a customer, interacts with an application.
For example, a customer wishes to return certain products and calls a customer service platform. They could speak to an AI vocal assistant who answers questions and requests as if they were talking with a human.
In a livestream, Openai T-Mobile customers have presented an AI engine agent who helps people find new phones. Another customer, the Zillow real estate research platform, presented an agent who helps someone refine a neighborhood to find the ideal place.
Openai said GPT-Realtime is its “most advanced vocal model ready for production”. Like his other vocal models, he can change my tongue in the middle of the sentence. However, Openai researchers noted that GPT-Realtime can follow more complex instructions such as “categorically speak in French accent”.
But GPT-Realtime faces competition from other models that many brands already use. Elevenlabs published the AI 2.0 conversation in May. Soundhound is associated with fast food franchises for an AI steering wheel service. The startup of Emphatic Hume AI launched its EVI 3 model, which allows users to generate AI in their own voice.
While companies discover various use cases for Voch IA, suppliers of even more general models that offer multimodal LLMs arise for themselves. Mistral has published its new VOXTRAL model, declaring that it would work well with real -time translation. Google improves its audio capacities and gain popularity with an audio functionality on notebooklm which converts search notes into podcast.
Better instruction following
Openai said GPT-Realtime is more intelligent and better understands native audio, including the ability to catch non-verbal signals such as laughter or sighs.
The comparative analysis using the Big Bench Audio Eval showed that the score model of 82.8% precision, compared to its previous model, which obtained 65.6%. Openai did not provide numbers testing GPT-Realtime against the models of its competitors.
OPENAI focused on improving the monitoring of the model instructions, ensuring that the model adheres to the directions more effectively. The new model obtains a score of 30.5% on the multichallenge audio reference. The engineers also strengthened the call function so that GPT-Realtime can access the right tools.
API updates in real time
To take charge of the new model and improve the way companies integrate AI’s capabilities in real time in their applications, Openai has added several new features to the API in real time.
He can now support MCP and recognize image entries, which allows him to inform users of what he sees in real time. This is a characteristic that Google strongly underlined during its presentation of the ASTRA project last year.
Real -time API can also manage the session initiation protocol (SIP). SIP connects applications to telephones such as a public telephony network or office phones, opening more cases of use of the contact center. Users can also record and reuse prompts on the API.
So far, people have been impressed by the model, although it is still initial tests of a recently published model.
OPENAI has reduced prices for GPT-Realtime by $ 20% to $ 32 per million audio entry tokens and $ 64 for audio output tokens.
https://venturebeat.com/wp-content/uploads/2025/08/crimedy7_illustration_of_a_half_machine_half_human_person_spe_63e21c7b-3093-4e77-b336-a83287f6af4a_2.png?w=1024?w=1200&strip=all
About The Author
