A new AI voice-cloning tool from a British firm claims to be able to reproduce a series of UK that he proclaims more accurately than some of its US and Chinese rivals.
Because most of the data used to traditionally train AI products come from North American or Southern English speaking sources, many artificial sounds make the same sound.
To combat this, the company synthesia spent a year, through recording people in the studio and collecting online content, compiling its database of UK sounds with regional accents.
It used those who are used to train a product called express-voyce, which can clone a real person’s voice or generate a synthetic voice.
They can be used in materials such as training videos, sales support and presentations.
The company said that its customers wanted more accurate regional representation.
“If you are the CEO of a company, or if you are just a regular person, when you have your equality, you want your pronunciation to be preserved.”
He said that French speaking customers had also commented that synthetic French voices inspired the French-Canadian sound rather than originating from France.
“This is just because companies that manufacture these models are North American companies, and they have datasets that are biased to demographics they are,” he said.
The most difficult accents for mimic are at least common, Mr. Mejjati said, as low recorded material is available to train the AI model.
There are also reports that voice-promoted AI products, such as smart speakers, are more likely to struggle to understand many accents.
Last year, concerns revealed in West Midlands Police internal documents Will the voice identity system understand the Broomi accent.
Meanwhile, the US-based start-up Sanas is taking opposite approach, developing equipment for deployment in call centers that “neutral” the accents of Indian and Filipino employees, As reported by Bloomberg in March.
The firm says that its purpose is to reduce the “pronunciation discrimination” experienced by the workers when the callers fail to understand them.
There is a concern that languages and dialects are being lost in the digital age.
“Of more than seven thousand languages who are still present today, almost half are in danger according to UNESCO; almost a third has some online appearance; less than 2 percent are supported by Google translation; and according to Openai’s own test, only fifteen, or 0.2 percent are supported by GPT-4. [an OpenAI model] Above 80 percent of accuracy, “Karen Hao writes in the AI’s book Empire.
AI expert Henry Ajder agrees, “Language models are giving speeches, which advise governments and technical firms including cinema.
However, the better these products are, the more effective they will be in the hands of the scammers.
The product of synthesia will not be free when released in the coming weeks, and the railings will be around abusive language and clear material.
But already there are many independent, open-sources voice-cloning tools that are easily accessible and less preserved.
In early July, the messages generated by AI-clone voice were reported to be sent to the US State Secretary Marco Rubio to the ministers.
“Open source landscape for voice has developed so fast in the last nine to 12 months,” says Mr. Ajder.
“And he, from a safety point of view, is a real concern.”
