Latest news with #syntheticvoice
BBC News
5 days ago
- Business
- BBC News
New AI voice tool trained to copy British regional accents
A new AI voice-cloning tool from a British firm claims to be able to reproduce a range of UK accents more accurately than some of its US and Chinese much of the data traditionally used to train AI products with voices comes from North American or southern English speaking sources, many artificial voices tend to sound combat this, the company Synthesia spent a year compiling its own database of UK voices with regional accents, through recording people in studios and gathering online used those to train a product called Express-Voice, which can clone a real person's voice or generate a synthetic can be used in content such as training videos, sales support and company said its customers wanted more accurate regional representations."If you're the CEO of a company, or if you're just a regular person, when you have your likeness, you want your accent to be preserved," said Synthesia Head of Research Youssef Alami added French-speaking customers had also commented that synthetic French voices tended to sound French-Canadian rather than originating from France."This is just because the companies building these models tend to be North American companies, and they tend to have datasets that are biased towards the demographics that they're in," he hardest accents to mimic are the least common, Mr Mejjati said, because there is less recorded material available to train an AI are also reports that voice-prompted AI products, such as smart speakers, are more likely to struggle to understand a range of year, internal documents from West Midlands Police revealed worries about whether voice recognition systems would understand Brummie the US-based start-up Sanas is taking the opposite approach, developing tools for deployment in call centres which "neutralise" the accents of Indian and Filipino staff, as reported by Bloomberg in March. The firm says it aims to reduce "accent discrimination" experienced by workers when callers fail to understand them. Endangered languages and dialects There is concern that languages and dialects are being lost in the digital era."Among the over seven thousand languages that still exist today, almost half are endangered according to UNESCO; about a third have some online presence; less than 2 percent are supported by Google Translate; and according to OpenAI's own testing, only fifteen, or 0.2 percent are supported by GPT-4 [an OpenAI model] above an 80 percent accuracy," writes Karen Hao in the book Empire of AI."Language models are homogenising speech," agrees AI expert Henry Ajder, who advises governments and tech firms, including the better these products become, the more effective they will also be in the hands of product will not be free when it is released in the coming weeks, and will have guardrails around hate speech and explicit there are already many free, open-source voice-cloning tools which are easily accessible and less the beginning of July, messages generated by an AI-cloned voice impersonating US Secretary of State Marco Rubio were reported to have been sent to ministers."The open source landscape for voice has evolved so rapidly over the last nine to 12 months," Mr Ajder adds."And that, from a safety perspective, is a real concern." Sign up for our Tech Decoded newsletter to follow the world's top tech stories and trends. Outside the UK? Sign up here.
Times
08-07-2025
- Times
His voice was taken by disease. Now he delivers justice through a synthesizer
A syrupy American voice congratulates members of the jury for being selected to serve in Dundee sheriff court. The inflection is almost celebratory, as though the line-up has secured new jobs or passed an exam. Although the words were supplied by Sheriff Alastair Carmichael, who has overseen proceedings in the city for 12 years, the voice was produced by Microsoft. 'You'll already have noticed that this is another synthetic voice that's speaking my words,' Carmichael's laptop tells the courtroom. Motor neurone disease (MND), which affects the nerve cells connecting muscles and the brain, has eroded his ability to enunciate words himself. The illness began with a 'numb, spongy feeling' inside his mouth in the autumn of 2023, then progressed to a lisp. Carmichael can still talk but the range of sounds demanded by the English language are no longer feasible. The letter 'C', he tells me, is particularly difficult. When I struggle to understand, as we chat during a morning in his chambers, he jots in a notebook or taps a phrase into his phone and shows me the screen. Bizarrely, during two holidays to France, he found French easier to enunciate. Of his diagnosis, he types: 'I'm not bitter about it. It is one of life's mysteries. You can only control what you can control.' Still agile, he moves nimbly around his book-lined room, providing refreshments and showing how he uses different digital devices. Carmichael's form of MND has only affected him from the throat up, a condition known as progressive bulbar palsy. His wife Helen, sons and courtroom colleagues who converse with him daily are much quicker at understanding his words than me. In order to do his job Carmichael uses a range of text-to-speech software and each programme has its quirks. One of 127 sheriffs in Scotland, Carmichael is thought to be the only judge in the UK, and possibly the world, presiding over cases using synthesised speech. 'Carrying on doing this gives me a purpose and enables me to be a full part of society by contributing,' he says. Carmichael recorded his own voice before he lost the power of speech. He had to read 300 sentences to create the necessary voice bank with SpeakUnique. As a result, his phone and PC can read his typed words in a tone his friends recognise. A phone app speeds up the process using text templates for common scenarios, such as shopping. Crucially for his work, the system is customisable and Carmichael has spent hours inputting the kind of phrases he is most likely to need in court. The MND team within NHS Tayside helped support this with a computer system called Grid 3. Press the tab for 'traffic offence' and it reads: 'On charge one you will be disqualified from driving for X months, reduced from X months because a plea of guilty means that a trial was not required.' Carmichael only needs to fill in the appropriate numbers in the courtroom on the day. He can also type during proceedings — he finds two fingers the fastest approach — swiftly granting two warrants for arrest on the morning I visit. He deploys the same technology for taking oaths. Translators, for example, have to promise to faithfully interpret during proceedings. Once, Carmichael says, he accidentally pressed the wrong key on his device. Instead of asking the interpreter to swear solemnly and sincerely he said: 'There is no alternative to a custodial sentence.' 'You have to retain a sense of humour,' he says with a smile. Carmichael comes from a family of engineers but took a different path because his maths was 'hopeless'. Before moving north he served as a High Court prosecutor in Edinburgh for seven years. Now living nearer to his wife's extended family, who farm, he says he does not miss life in the central belt. In 2023 he sentenced Tracie Currie and Carl O'Brien for targeting Humza Yousaf, then the first minister, with racist abuse. Last November he sentenced the Earl of Dundee, Alexander Scrymgeour, for drink driving. When hearings go to trial, the systems that use his synthetic voice cannot rise to the occasion, unable to handle text longer than three sides of A4. Carmichael calls his words for a trial on to the screen. With all his directions to jurors, it runs to page 18. For this to be heard he relies on Microsoft Word, which cannot use his voice and instead provides its own. This is why his opening remarks to the jury are delivered in an American drawl. 'I cannot get rid of it,' he says. He can select the gender of the speaker and the system offers English narrators known as Hazel and George, but Carmichael says he cannot always control who shows up to the courtroom. He demonstrates a section of text delivered in a more soothing lilt, known as 'smooth' George, although Carmichael is not sure why this virtual character takes over his monologue at this point. 'Sometimes it is a complete surprise to me which voice comes to the microphone,' he says. There are pros and cons to this technological uncertainty. Carmichael emphasises the importance of the jury trusting him, but he also sees the possibility of a sudden shift in voice keeping the 15 men and women engaged. 'My laptop becomes a point of interest, who knows which voice might pop up next,' he writes. The Scottish Courts and Tribunal Service is working on a solution that will allow his own synthesised tones to be used more extensively. Carmichael has handed out hundreds of criminal sentences using voice technology and since the system was launched for jury trials last November, after a period of testing, he has adjudicated in a dozen jury trials. There have been no complaints thus far. People, he notes, are well accustomed to technology. It is the jar filled with slips of paper for picking jurors' names I find anachronistic, not his laptop on the bench. 'The important thing is [that] as long as the words are my words, an objection will not succeed,' he explains. 'For example, if I was using artificial intelligence that would be a bad thing, but I am not. I'm always making sure it is what I want to say before I say it.' The harder it has become to speak, the less self-conscious he has felt about relying on all the other options, he says, writing down 'self-conscious' because it is hard to mouth. 'I think you cannot really understand unless you have experienced something similar,' he continues. 'It is also quite humbling. I am in a new situation where I am more reliant on other people making allowances and adjusting what they do in order to accommodate me.' He says the hardest thing to deal with in court is when a witness is prevaricating or behaving offensively. 'Then you have to type things, but I cannot nuance. You have to just say, 'Answer the question'.' If someone becomes upset on the stand, he always uses his recorded voice to help them calm down, as it 'sounds more empathetic'. Carmichael does ponder how important one's voice is to personality. Aspects of communication he misses include pausing when he would like, making eye contact and gesturing as he talks, which feel absent. The emphasis of repetition in normal speech patterns is also gone. But he has learned to add extra commas to create a more natural sound and misspell some words so they are pronounced correctly. 'The systems don't like Scottish, or dialect words, and many of them get a verbal mangling unless I misspell them,' he explains. The Aberdeenshire village of Strachan is one example, which will be pronounced with a soft 'ch' in the middle unless he writes 'Stracken'. Spontaneity, Carmichael says, is what he misses the most. 'I think of something I want to say but by the time I have put it in my phone or written it down, the conversation has moved on.' Sometimes in meetings he raises his hand to indicate he has a contribution. Backed by his boss, Sheriff Principal Gillian Wade, his approach to each challenge is to simply crack on. MND, though incurable, affects patients so differently that his prognosis is unknown. He feels well. He is aiming to reduce his 'very average' golf handicap before he is 'physically unable' to play. For now, he can eat everything he wants, although it 'takes a lot of time' and a cough or sneeze 'is like a car wash'. While losing the ability to swallow is a worry, he has determined not to let fear dominate. 'I am not going to waste time and energy being miserable,' he says.
