Imagine this: You work in the accounts department of a large corporation. Your manager calls you and asks you to transfer money to a specific account at short notice or a loved one is in trouble and needs money for an emergency such as a car accident or hospital bill. Would you make the transfer or provide the money? We hope not directly, because in reality, the person on the phone could be a stranger using AI voice cloning software.
What is AI voice cloning?
AI voice cloning technology uses deep learning algorithms to analyse and reconstruct a person’s voice. While this used to require large amounts of data from voice recordings to recognise the person’s speech pattern and create a synthetic voice that is similar to the person’s natural voice, today it takes less than a minute of voice recording.
How does AI voice cloning work?
Deepfake technology analyses everything that makes a person’s voice unique, including age, gender and accent. It can then recreate the pitch, timbre and individual sounds of the voice to achieve a similar overall effect.
All you need is a short voice sample of the human voice, such as posts on YouTube, TikTok, Instagram or Facebook videos, and the AI learns it immediately. You can then write a script of your choice, which the AI then reads out in the modelled voice. Now you might be thinking: ‘My voice can’t be found publicly on the internet, now I’m safe!’ Unfortunately, it’s not quite that simple. A fake call to you (which is recorded by the attacker) can be enough to capture enough voice material.
Dangers of AI voice cloning
Of course, this technology should not be seen in a one-sided light and offers people who have difficulty speaking in particular a synthetic voice to express their thoughts. However, the potential for abuse is huge and ranges from false statements that jeopardise reputations to sophisticated scams involving family emergencies and CEO fraud.
This type of attack doesn’t stop at money either. Such services only cost a few dollars a month and can often be tested free of charge. There are also more and more open-source projects in this area that can be used by attackers free of charge.
Voice cloning enables users to recreate a person’s unique voice patterns. This means that our voice will soon be as worthy of protection as our fingerprints or DNA. Of course, this also gives rise to the following legal issues.
3 ways to protect yourself from voice cloning fraudsters
Developments in the field of AI and voice cloning are currently progressing very quickly. Voice models are getting better and better, and the number of voice templates required is constantly decreasing. There are various ways to protect yourself from the dangers of AI voice cloning:
1. Awareness: low-tech methods to combat a high-tech problem.
There is no technical way to defend against such attacks, only creating awareness of such attacks can help to avoid becoming a victim yourself. Here are a few practical tips that you can follow to recognise a suspicious call, e.g. from a supposed boss:
- Is there an attempt to build up pressure – e.g. time/appointment pressure?
- Is the manner of speaking unusual – e.g. are the greetings different from usual?
- Are the answer sentences shorter than usual or do they not exactly match your question?
- If one or more of these points apply, it is quite possible that it is a fake call.
- Double check: Check the identity of the caller
- End the call and call the person back. If the attacker has used a fake telephone number, you will not get through to the attacker, but to the real person and the attack will be averted!
2. Double check: Check the identity of the caller
End the call and call the person back. If the attacker has used a fake telephone number, you will not get through to the attacker, but to the real person and the attack will be averted!
3. Minimalism!
Be aware that making your voice public can be used against you and makes it easier for potential attackers.
Conclusion
AI voice cloning has a high potential for fraud. This can be attributed to two factors in particular:
Ease of access:
Potential perpetrators hardly need any technical knowledge to use AI Voice Cloning for their purposes
Favourable price:
Potential attackers do not have to dig deep into their pockets to use voice cloning technology, if at all.
These two factors lead to very low barriers to utilisation. Although there are initial recognition technologies for cloned voices, the most effective protection remains: security awareness. And this applies to both private and corporate environments.