A new AI voice-cloning tool developed by British company Synthesia claims to replicate a broad range of UK regional accents more precisely than some of its American and Chinese competitors.
Traditionally, most voice data used to train AI systems comes from North American or southern English speakers, resulting in synthetic voices that often sound similar and lack regional variety. Reports Technology News
To tackle this, Synthesia spent a year building its own database of UK voices with distinct regional accents. The team recorded people in studios and gathered audio from online sources to create a more diverse dataset.
This database was used to train a tool called Express-Voice, which can either clone a real person’s voice or generate synthetic ones. The voices can be used in training videos, sales materials, and presentations.
Synthesia says there’s growing demand from customers for more accurate regional representation in AI voices.
“If you’re the CEO of a company, or just a regular person, you want your likeness to include your accent,” said Youssef Alami Mejjati, Head of Research at Synthesia.
He added that French-speaking clients had also raised concerns, noting that AI-generated French voices often sound more like French-Canadian speakers than those from France.
“This is because most of the companies developing these models are based in North America, and their datasets reflect the demographics around them,” Mejjati explained.
The most difficult accents to replicate, he said, are those that are least commonly heard — simply because there’s not enough recorded data available to train the models effectively.
Voice-activated AI products, such as smart speakers, also tend to struggle with understanding less familiar accents. Last year, internal documents from West Midlands Police highlighted concerns that voice recognition systems might not accurately understand Brummie accents.
In contrast, US-based start-up Sanas is taking a different route. It has developed technology for call centres that “neutralises” the accents of Indian and Filipino workers. According to a Bloomberg report from March, the company says this helps reduce “accent discrimination” — a common problem when customers fail to understand workers’ speech.
