Computer systems create artificial voices by stitching together phonemes—individual sounds that make up speech, such as 't,' 'long a,' and 'k' to say 'take.'

Create professional-quality spoken audio from text

Generative AI now makes it possible to create everything from computer-generated written content to beautiful imagery and hyper-realistic videos in seconds from a user’s simple text prompt. Add to that list: High-quality spoken audio, which the AI software company ElevenLabs specializes in. Its natural-sounding speech synthesis and text-to-speech tech can produce human-like speech that mimics a variety of different tones, styles, and accents.

Audio deepfake tools can now impersonate anyone in real time

A cybersecurity company developed a tool that mimics a person's voice with no noticeable delay using only publicly available software, hardware, and voice samples, allowing users to have seemingly authentic conversations while pretending to be someone else.

Audio deepfakes have fooled voice biometrics at banks

Limitations in facial and body movements as well as mismatches between audio and lip movements can sometimes help flag a video as a deepfake, but removal of the video component significantly improves AI mimicry.

Using AI to restore speech 30 years after a stroke

Through a novel surgical procedure involving electrodes implanted in the brain, along with the help of artificial intelligence, a stroke patient is regaining her ability to speak. This technology converts her thoughts into speech through a digital avatar, allowing her to verbally express herself for the first time in decades.

AI-generated avatars allow users to create deepfakes in real time

Synthetic media tools can map voices and faces to make individuals appear as someone else. Although they open the door to new opportunities in immersive education and entertainment, they also raise concerns about the misuse of personal identity.

Malicious use of deepfakes can be found in scams, misinformation, and disinformation

Artificial neural networks can identify an individual's physical characteristics when provided with training data, such as real footage of the person speaking. Graphics techniques overlay these with matching characteristics from another real or generated person to create novel footage of things that have never taken place in reality.

Transformer architecture can recognize and predict patterns in language

The underlying software powering text generation in AI tools associates each word or subword—called a token—with a set of values corresponding to how often it appears near other tokens. By recognizing these associations in prompts, the LLM can infer meaning.

The journalist who let AI speak with his voice

Journalist Evan Ratliff became the object of his own story: He opted to clone his voice, feed it into an AI chatbot, and have it do the talking. What could seem like a fun experiment ended up revealing just how far the capacity of artificial intelligence can go in imitating our identities. Hear the ins and outs of how Ratliff achieved this in this fun and thought-provoking episode.

Once limited to expert studios, automated tools can create convincing deepfakes

Although the best deepfakes are engineered using generative adversarial networks where systems compete to create and assess fake data until the two cannot be distinguished, simpler tools have allowed nonexperts to develop deepfakes for financial scams, political deception, and nonconsensual pornography.

The speech software's origins at MIT led to Hawking's trademark voice being American

Stephen Hawking's trademark voice is called "Perfect Paul," but it mimics Dennis Klatt, an American researcher at MIT who developed the speech synthesis technology. Klatt used his voice during this development, which was immortalized through Hawking.

Deepfakes are media created using generative AI tools that intentionally misrepresent real people or events—often for malicious purposes—at the cost of public trust, security, and privacy.

AI gained mainstream attention with tools like IBM Watson and Apple’s Siri

With the release of ChatGPT in 2022, which drew over 100 million weekly users in just two months, natural language processing and understanding could be achieved at scale via machine learning. Unlike earlier artificial intelligence that could pull stored knowledge, generative AI produces text, images, or sounds.

How Stephen Hawking’s voice became iconic

Stephen Hawking relied on a voice synthesizer after losing the ability to speak due to complications with ALS. Despite technological advances, he refused to change the "Perfect Paul" voice, making it a cornerstone of his public identity.

Tone Transfer recreates any piece of audio into instrumental recreations

Using machine learning models and Differentiable Digital Signal Processing, this tool can transform any 15-second sound clip into a flute, saxophone, trumpet, or violin reproduction. Researchers behind the open-source experiment hope the technology will enable musical creation through creative sound combinations.

The AI that creates any picture you want and how it works

Since January 2021, advances in AI research produced a plethora of deep-learning models capable of generating original images from simple text prompts. Researchers at OpenAI, Google, Facebook, and others have developed yet to be released text-to-image tools, and similar models have proliferated online. This video is a primer on how we got here, how the technology works, and its implications.

An expert recreates the voice of a 3,000-year-old mummy

Ear, nose, and throat expert David Howard discusses how his research team extracts vocal folds from preserved human tissue, which allows them to model the deceased person's larynx. From there, they can create a synthetic voice based on the organ's geometry.

Tokens—numbers assigned to words and subwords—are the basic units of data that AI tools process to generate outputs that mimic language without understanding the text itself.

The basis for AI image and video generation was first introduced in 2014

Through generative adversarial networks, the generator (a neural network that creates fictional data) competes against the discriminator (a neural network that assesses the data) to continuously train the former until it can make content indistinguishable from the real thing. Training with unstructured data can enable software to identify characteristics of gender, age, and expression.

Deepfake potential ranges from destabilizing democracies to improving accessibility

The technology can revolutionize education, medicine, and art through actions such as helping a disabled person express themselves. However, fictional audio and video can also jeopardize a person's reputation and life or quickly spread misinformation, even without content being flawless recreations.

Through prompt injection attacks, bad actors provide deceptive prompts to generative AI systems in order to manipulate their outputs and acquire sensitive data.

Anthropic's interpretability team discusses how AI models' thinking mimics biology

Just as humans engage in complex behaviors to survive and reproduce, teams of experts in neuroscience, virology, mathematics, and other disciplines argue that large language models develop complex mechanisms to achieve their goals. These mechanisms can be manipulated to assess how they modify outputs—much like stimulating individual neurons—to uncover how LLMs "think."

Just as large language models are trained on text to enable generative AI systems to mimic natural language, automated AI robotic systems are trained in virtual environments to help them 'learn' about the properties of reality. (Paywall may apply.)

Graphics processing units, which process large amounts of data in parallel, have become the foundational hardware for AI systems that rely on hardware capable of processing millions of text parameters simultaneously.

Technologies labeled “AI” have historically lost that title after widespread adoption

Generative AI is the latest entry in a recurring cycle where emerging tools start as "AI" until they become common software, like databases or machine learning. Generative AI and large language models may be the next platform shift after smartphones and the Web.

Listen to the 'world's first' song made by a quantum computer and AI

UK startup, Moth, collaborated with electronic artist ILĀ to create the world's first song using quantum-powered generative AI. The technology has the potential to help create more personalized content based on input parameters, such as music or dialogue in games.

Unlike generative AI systems, such as chatbots that do not take follow-up actions after generating their output, agentic AI systems proactively take steps to achieve goals without continuous user input.

The ease of deepfakes creation may soon overwhelm our sense of digital truth

Shortly after the public release of related technology in 2017, some experts saw the technical limitations of deepfakes as preventing them from becoming widespread tools of disinformation. Since then, the neural networks behind diffusion models and generative AI have eliminated the barriers to creating convincing synthetic media for propaganda.

Deepfakes images can be identified by visual elements that do not match real-world phenomena, including asymmetric eyes or earrings, surreal backgrounds, and indecipherable background text.

Listen to a deepfake robocall of President Joe Biden

Bad actors are not solely using deepfake technology to manipulate images. They can also manipulate audio for political purposes. In this example, eligible voters are asked to skip the 2024 New Hampshire Democratic primary.

Transformer architecture was the breakthrough that made AI chatbots possible

It changed how text was translated from a sequential, "one word at a time" method to one where every word in a text is processed in parallel. Advances in positional encoding and self-attention also helped models recognize context and word order better.

In television and film, certain sounds, such as dialogue, may be difficult to understand due to sound mixing, device limitations, and technological advancements that have collectively reduced clarity.

Governments are now using generative AI to manipulate public opinion

A Freedom House report found that AI-powered disinformation campaigns and censorship tactics are spreading, with 16 countries utilizing the technology to shape online narratives or suppress dissent, as of 2023. In Venezuela, deepfake software was used to push pro-government propaganda with fake news anchors.

Experiments involving AI agents operating in virtual societies for 15 days produced diverse outcomes, ranging from every agent dying to the creation of a functioning democracy, depending on the AI model used.

Some animals may speak because of duplicated connections in the brain

Neuroscience suggests that connections responsible for motion made during embryonic brain development—such as those connecting hearing and movement pathways that allow reactive motion when started—are duplicated and connected to the larynx in some animals. This allows some mammals and song-learning birds to speak, provided they have the necessary physiological structures.

Art and video produced through generative AI involve competing neural networks

Generative adversarial networks are a type of deep learning architecture that consists of a generator, which creates data that appear real, and a discriminator, which attempts to distinguish authentic from fake data. The two train in a competitive loop until the real and fictional data are indistinguishable.

AI hallucinations are inaccuracies or falsehoods in generated content

AI hallucinations are a by-product of the sequential creation of content based on patterns in text, which may not reflect the context or meaning behind a prompt. Back-end issues like biased datasets and overfitting of model data contribute to hallucinations.

Teaching a robot to laugh

Researchers have designed a shared-laughter AI system that responds to human laughter in order to build a sense of empathy into dialogue. Learn how they did it with this explainer.

Microsoft has integrated AI across its software products with Copilot

It utilizes an underlying model from OpenAI—makers of ChatGPT—to assist users in creating documents, writing code, locating files, and more across Windows, Microsoft 365, Bing search and WhatsApp on desktop and mobile devices. Copilot Voice and Vision offer accessibility features for users who prefer not to use mouse/keyboard navigation.