AI Voice and The Dangers of Cloning

The Voice of the Future: Can AI Save Our Accents, or Erase Them?

Artificial intelligence now possesses a voice. In fact, it possesses millions of them. Groundbreaking technology allows companies to clone, create, and modify human speech with astonishing realism. For some, this heralds a new era of personalised digital communication and cultural preservation. For others, it is a dangerous step towards a homogenized world and a powerful new tool for criminals. This technology's dual-power to both replicate and erase our vocal identity places society at a critical crossroads. A company from Britain is pioneering the accurate digital preservation of regional accents, while a US counterpart is offering to strip them away. The very essence of how we sound is now programmable, raising urgent questions about identity, security, and what it means to speak in the 21st century.

A Quest for Vocal Authenticity

A technology firm in Britain has embarked on a mission to solve a persistent problem in artificial intelligence. Synthesia, the company in question, claims its new AI system is capable of replicating a broad spectrum of UK pronunciations with more accuracy than its American or Chinese competitors. The project addresses a significant bias in the world of AI voice generation. Historically, the training information for these systems has overwhelmingly been sourced from speakers in North America or the south of England. This has led to a landscape where a large number of synthetic voices have a strikingly similar sound, lacking the rich diversity of genuine human speech and alienating users who do not fit that narrow vocal mould.

Building a British Voice Bank

To counteract this data bias, Synthesia invested a year in a unique and ambitious project. The company set out to build its own bespoke database composed entirely of British voices, capturing the nuances of regional accents from across the United Kingdom. This extensive undertaking involved recording a wide variety of people in professional studio environments. The firm also gathered a substantial amount of material from online sources to broaden the dataset. The result of this year-long effort is a rich and varied library of vocal data, forming the foundation of their innovative new product. This database represents a significant step forward in creating more inclusive and representative AI.

The Express-Voice Engine

With its curated voice bank, the company developed Express-Voice, a new system. This powerful tool has a dual function. It can generate entirely new synthetic voices, or it can make an exact digital copy of an individual’s speech patterns. The primary applications for this technology are in the corporate and commercial sectors. Businesses can use these authentic-sounding voices in a wide range of content. This includes materials for employee instruction, assistance for sales teams, and external presentations. The aim is to provide a more genuine and relatable experience for audiences, moving away from the generic, robotic narration that has long characterised text-to-speech systems.

Preserving Personal Identity

The driving force behind this innovation was clear demand from the market. According to Synthesia, its customers were increasingly requesting truer local representation in their digital content. The head of research for the company, Youssef Alami Mejjati, explained that accent is a core part of personal identity. He stated that whether you are a chief executive or an ordinary employee, an individual’s expectation is for their digital representation to maintain their distinct accent. This desire for authenticity is not limited to English speakers. Mejjati noted that French-speaking clients had also complained about synthetic voices having a French-Canadian sound instead of a European French one, a direct result of the geographic bias of the companies in North America that build them.

Image Credit - Freepik

The Accent Recognition Gap

The challenge of accurately mimicking accents highlights a broader issue in AI development. Accents that are not widespread are invariably the most difficult to replicate. Mr Mejjati explained this is due to a simple lack of available data. When an artificial intelligence system has a smaller quantity of documented material to learn from, its ability to generate a convincing replica is significantly diminished. This problem extends beyond voice cloning and affects many popular AI products. There are widespread reports of voice-prompted smart speakers and digital assistants experiencing difficulty comprehending different regional accents. This digital communication barrier reveals the inherent biases baked into the systems many people use every day.

Is Anyone There? AI's Regional Blind Spot

Studies confirm the frustration felt by many users of AI assistants. Research by the consumer advice firm Uswitch found that voice assistants from major tech giants have difficulty understanding nearly a quarter of all regional accents in the United Kingdom. The study involved recording people from 30 different cities asking their devices a series of basic questions. The results showed a clear geographical disparity. Voice assistants struggled most significantly with accents from Cardiff, Glasgow, and Liverpool. In contrast, London, Lincoln and Chester were the most easily understood locations, suggesting the AI models were trained on data that reflects a southern English standard.

Police Concerns and Public Scepticism

This technological shortcoming has real-world implications. Official worries came to light from West Midlands Police in internal papers last year about whether new systems for voice recognition could comprehend Brummie accents. The issue is not just one of convenience; it touches on accessibility and equal access to technology. Further research from Abertay University has raised alarms about how the increasing realism of AI voices could be exploited. A recent study highlighted that people, particularly those from Scotland, had a higher tendency to perceive an AI-generated voice with a regional Dundonian dialect as human, creating a potential opening for scammers. This suggests that familiarity can breed a false sense of security.

The Opposite Approach: Accent Neutralisation

While one company works to preserve accents, another is focused on removing them. The US-based startup Sanas is developing and deploying tools for call centres that make the pronunciations of their staff neutral in real-time. The technology targets the voices of workers in countries like India and the Philippines, altering their speech to sound more like a generic Western accent. The company's stated aim is to reduce the "accent discrimination" that employees can face when callers have difficulty comprehending the workers or react with prejudice. The firm argues that its technology improves communication clarity and agent well-being.

Image Credit - Freepik

A Booming, Controversial Business

Sanas’s approach has proven to be extremely lucrative. The company has attracted significant investment, recently closing a $65 million funding round that values it at over $500 million. It has acquired a competitor and reports that its technology is already in use at dozens of companies, including some of the world's largest business process outsourcing (BPO) firms. This rapid growth indicates a strong market demand for accent modification. However, the technology operates in a deeply controversial space. Critics argue that such tools risk homogenising voices, erasing cultural diversity, and pandering to, rather than challenging, societal biases.

An Ethical and Cultural Minefield

The debate around accent-softening AI is complex. While Sanas frames its mission as one of breaking down barriers and reducing discrimination, some see it as a form of digital "whitewashing." They argue that the technology reinforces the idea that non-Western accents are a problem to be solved, rather than promoting greater exposure and acceptance of linguistic diversity. The very idea of a "neutral" accent is considered by many linguists to be an ideological construct rooted in social hierarchies. This places Sanas at the centre of an ethical storm, with its success highlighting a corporate willingness to erase cultural identity for the sake of smoother customer interactions.

The Battle for the Voice Market

The burgeoning market for voice modification technology has become fiercely competitive, even leading to legal disputes. In a recent development, Sanas filed a lawsuit against a rival company, Krisp, in July 2025. Sanas alleges that Krisp, after expressing interest in licensing Sanas's technology and signing a non-disclosure agreement, developed a "copycat" version of its accent translation software. The lawsuit claims theft of intellectual property and trade secrets. This legal battle underscores the immense commercial value now placed on AI voice manipulation and the high stakes involved for the companies racing to dominate this emerging sector.

The Dark Side: A New Wave of Scams

As the technology becomes more sophisticated and accessible, it also becomes a more effective weapon for criminals. Law enforcement agencies across the globe have issued stark warnings about the rise of scams utilising AI voice cloning. Fraudsters can now create a convincing audio replica of someone's voice with just a few seconds of sample audio, often scraped from videos posted on social media. According to research from Starling Bank, an alarming 28% of UK adults believe they have already been targeted by an AI voice-cloning scam, with many unaware the threat even exists. The age of audio deepfakes is here, and it is fuelling a new generation of fraud.

The ‘Grandparent Scam’ Goes High-Tech

One of the most common and cruel applications of this technology is the so-called "grandparent scam." Criminals use a cloned voice of a grandchild to call an elderly relative, feigning distress and claiming to be in trouble. They might say they have been arrested or are stranded abroad and need money urgently. The emotional distress and the shocking realism of the voice can easily overwhelm a victim's judgement. The FBI has warned that this method is becoming increasingly prevalent, preying on the love and protective instincts of senior citizens, a group that lost an estimated $3.4 billion to financial crimes in 2023.

Image Credit - Freepik

When Hearing is No Longer Believing

The threat extends far beyond family emergencies. In a high-profile case from February 2024, an employee at the engineering firm Arup was tricked into wiring $25 million to fraudsters. The criminals used deepfake technology to stage a video conference call that included a digitally recreated, voice-cloned version of the company’s chief financial officer. This type of sophisticated fraud, known as vishing (voice phishing), is on the rise. Cyber security firm CrowdStrike reported a staggering 442% increase in the use of AI voice cloning between the first and second half of 2024, highlighting how quickly criminals are adopting these tools.

The Security Response Challenge

This new reality poses a significant challenge for security systems that many people and institutions rely on. For years, banks and other organisations have used voice verification as a supposedly secure method of authentication. A BBC investigation in late 2024 demonstrated just how vulnerable these systems now are. A journalist successfully bypassed the voice ID security of two major UK banks using a commercially available instrument for AI voice cloning. This has prompted warnings from technology leaders, including OpenAI, urging businesses to phase out voice-based authentication as a primary security measure, as it may now be obsolete.

A Digital Lifeline for Dying Languages

Amid the alarm bells, there is a more hopeful story. The same underlying AI technology that fuels these scams also holds the potential to become a powerful tool for cultural preservation. Data from UNESCO indicates that of the more than 7,000 languages spoken today, nearly fifty percent face the threat of extinction. For centuries, globalisation and the dominance of major languages have pushed minority tongues towards extinction. Now, AI offers a digital lifeline. Its application can help create vast archives of endangered languages, processing and transcribing oral histories and conversations with unprecedented speed and accuracy, preserving them for future generations.

AI in the Field of Language Revitalisation

Around the world, pioneering projects are already demonstrating AI's potential. In New Zealand, Te Hiku Media has used artificial intelligence to create a speech recognition tool for Te Reo Māori, which can transcribe the language with 92% accuracy. In Bangladesh, the International Mother Language Institute is using AI to document indigenous languages and has even been instrumental in preserving languages that were already considered extinct. Furthermore, initiatives like Mozilla's Common Voice are building open-source datasets for a huge number of languages, providing the raw material needed to build more inclusive AI tools and language-learning platforms.

Image Credit - Freepik

The Global Mission for Linguistic Diversity

These efforts are part of a global movement to protect our shared linguistic heritage. UNESCO is leading the International Decade of Indigenous Languages (2022-2032) to raise awareness and mobilise resources for the preservation, revitalisation, and promotion of these vulnerable tongues. AI-powered chatbots and virtual tutors can create interactive and accessible learning environments, helping to pass these languages on to a new generation of speakers. By documenting unique grammatical structures and vocabularies, AI is not just saving words; it is helping to preserve the cultural identities, traditional knowledge, and unique worldviews embedded within them.

Navigating the Ethical Maze

The rapid advancement of voice cloning demands a robust ethical framework to guide its use. A core principle must be informed consent. Companies developing this technology have a responsibility to be transparent about how they collect, use, and protect voice data. Ethical guidelines published by organisations like Synthesia and AudioStack stress that an individual's voice should never be cloned without their explicit, informed, and revocable permission. This includes establishing clear ownership of the resulting voice clone and implementing stringent security measures to prevent its misuse. Regular audits and clear content moderation policies are essential to build trust.

The UK's Regulatory Landscape

In the United Kingdom, the use of AI voice cloning falls under several existing legal frameworks. The collection and processing of voice recordings, which can be considered personal data, are governed by the Data Protection Act 2018 and the UK's version of the GDPR. This legislation requires a valid legal basis for processing such data and mandates that individuals are informed about how their information will be used. While these laws provide a foundation, the specifics of AI-driven technology are constantly evolving, creating a need for continuous review and potential new legislation to address the unique challenges that voice cloning presents.

A Future of Synthesised Speech

Society stands on the brink of a new audio era, one where the distinction between human and synthesised speech is increasingly blurred. AI voice technology offers incredible opportunities, from creating more engaging and accessible digital experiences to preserving the linguistic diversity that defines our shared human culture. Yet, it also opens a Pandora's box of potential misuse, from sophisticated financial scams to the erosion of personal and cultural identity. The path forward requires a delicate balance of innovation, regulation, and public awareness. The voice of the future is being built today, and the choices we make now will determine whether it speaks with a chorus of diverse accents or a single, synthesised tone.