A year later, OpenAI still hasn’t released its voice cloning tool

In the end of March last March, Opena announced an “small -scale preview” of an AI service, Voice engineThe company has claimed that a person’s voice can clone with just 15 seconds of speech. About a year later, the equipment remains preferably and does not indicate when the opeina may be launched – or whether it will be launched at all.

The company may point out the fear of abuse of the company to make the service greatly, but it may also reflect an attempt to avoid the invitation of the regulatory investigation. Openai Histor is tihassically Been charged Priority expenditure and “shiny products” prioritized and Rushing release To defeat the rival companies in the market.

In a statement, a spokesman for the OpenAI told TechCrunch that the company was experimenting with a limited set of “trusted partners” with a limited set of voice engines.

“[We’re] How to learning from [our partners are] Using the technology so that we can improve the effectiveness and protection of the model, “the spokesperson said.” From speech therapy to language learning from language to customer support, video game characters, the way II incarnation is being used. “

Pushed back

Voice Engine, which gives the Open-to-Text-to-Spit API as well as ChatzPT’s Available Voices Available Voice Voice modeNatural-sounding produces speech that is similar to the original speaker. The equipment converts the written letters into a speech, limited to the specific maintenance in the content. However, it was subjected to delay and release window from the beginning.

OpenAI explained in June 2024 Blog postThe voice engine model has learned to predict the most possible words for the transcript given by a speaker on a speaker, taking into account different voice, accents and speaker styles. After that, the model can not only create the spoken versions of the text, but also create “spoken words” which reflects how different speakers read the text aloud.

According to a draft blog post seen on TechCrunch, Open was initially intended to bring a voice engine, originally known as custom voice. The plan was to provide a “social facility” to a team of access to “trusted developers” up to 100 before a greater debut, or shows the use of “innovative and responsible” of technology. Opena was even Trademark And the price determines: “Standard” voice is $ per million dollars and “HD quality” Voice per million letters per million letters.

Then, the eleventh hour, the company suspended the announcement. Opena finished the voice engine unveiled without sign-up options a few weeks later. OpenAE said that access to the equipment will be limited to a association of about 10 Daves, the company has started work in the end of 2023, Openai.

“We hope to start a conversation on how synthetic voices and society can adapt to these new powers,” Open A Voice Engine’s announcement wrote in the blog post By the end of March 2021. “Based on the results of this conversation and the results of this small size, we will make more informed decisions on how and how to put this technology on a scale.”

Work

According to the Openai, the voice engine has been working since 2022. Organization Claim It has demo the “Global Policymakers at the highest level” in the summer to display its potential – and the risks.

Several partners have access to voice engine today, including startup levox, creating devices that enable disabled people to communicate more naturally. CEO Carlos Pereira told TechCrunch when Livox eventually did not create a voice engine on a product due to the online requirement of the equipment (many customers in Livox), he considered the technology “really impressive”.

Perira told Techchen through email, “Voice quality and the possibility of having voices in different languages ​​is unique – especially for people with disabilities,” our customers, ““ This is really impressive and easily usable. [tool to] Create the voice I have seen […] We hope that Open is soon develops an offline version. “

Pereira says he did not receive a guidance from the Opeena at a potential voice engine launch, nor did he see any sign of planning to start charging for the service of the company. So far, Livox did not have to pay for its use.

In June 2021, the aforementioned openly indicated that one of the considerations in the delay in voice engine was likely to abuse during the US election cycle last year. Talks with stakeholders are aware of the audio -generated audio of the voice engine, including watermarking, there are several miraculous protection systems.

According to the opening, the developers must accept “clear consent” from the original speaker before using the voice engine, and they must “clearly reveal” their audience that the voices are AI-exposed. But the company did not say how these policies are implementing. Working on a scale for any organization, including OPENCIES, can be proven to be extremely challenging.

Opina also refers to its blog posts that it hoped to create a “voice authentication experience” to verify the speakers, and a “non-go” list that prevents a very similar voice with prominent personality. Both are technically ambitious projects and these are wrongly reflected on any organization that is often accused of Removing the security initiativeThe

Effective filtering and ID verification is becoming a baseline requirement for rapid voice cloning tech release. AI Voice Cloning was the third fast -growing scandal of 2024, According to a sourceThe It led Fraud And Bank security check Privacy and bypassing as copyright laws are struggling to continue. Malicious actors have used voice cloning to create infection dipfkes Celebrity And PoliticiansAnd there is that depth Fire Across social media.

Open can publish the voice engine next week – or never. The company has repeatedly said that it is weighing to keep the service smaller. But one thing is clear: Due to optics, the reasons for safety or both have become one of the longest in the history of the open -end openness of the voice engine.