The language of deepfakes

AI generated Speech

Any audio output produced by a machine learning speech synthesis model.

AI generated speech encompasses all audio output from machine learning models trained to produce human-like speech. This includes text to speech (TTS), voice conversion, and zero shot voice cloning systems. AI generated speech is the raw material of audio deepfakes. When used to impersonate real people without consent, it becomes a security and integrity threat that requires detection tools like Raid AI.

AI generated Threats

Cybersecurity threats that rely on generative AI to attack targets.

AI generated threats are a category of cybersecurity risks in which the attack itself is produced by artificial intelligence. Examples include AI crafted phishing emails, automated vulnerability discovery, deepfake audio used for voice fraud, synthetic face images for identity forgery, and AI generated video used for disinformation and impersonation. These threats scale faster, cost less to produce, and are harder to detect than traditional human-crafted attacks. Defending against them requires AI native tooling such as Raid AI's deepfake detection platform, which analyses audio, image, and video content.

Audio Forensics

Scientific examination of audio evidence for authenticity and origin.

Audio forensics is the scientific discipline of analysing recorded audio to determine authenticity, identify speakers, detect tampering, or recover degraded content. Traditional audio forensics examines noise floor, compression artifacts, and splicing points. Modern audio forensics must also detect AI generated content and deepfake insertions, which require specialized machine learning models beyond classical signal processing techniques.

Audio Liveness Detection

Verification that audio comes from a live human rather than a replay or synthesis.

Audio liveness detection is the process of confirming that a voice sample originated from a real, present human speaker, rather than a recording, replay attack, or AI generated synthesis. Liveness detection is essential for high-security voice authentication systems in banking and government. Raid AI provides a modern, AI first approach to audio liveness detection that is effective against the latest generation of voice cloning tools.

CEO Fraud

A fraud scheme that impersonates a company executive to authorise payments.

CEO fraud, also called business email compromise (BEC) or whaling, is a scheme where attackers impersonate a senior executive to trick employees into wiring money or disclosing sensitive data. Modern CEO fraud increasingly uses deepfake audio, with attackers cloning a CEO's voice and calling finance staff to 'authorise' urgent transfers. Individual deepfake CEO fraud incidents have resulted in losses exceeding $35 million.

Deepfake

AI generated media made to impersonate a real person.

A deepfake is synthetic audio, image, or video content generated by artificial intelligence that imitates the appearance or voice of a real person. The term combines 'deep learning' and 'fake'. Audio deepfakes recreate a target speaker's voice, accent, and speech patterns so convincingly that listeners cannot reliably distinguish them from authentic recordings. Image and video deepfakes swap faces, generate synthetic likenesses, or manipulate visual content to fabricate scenes that never occurred. Raid AI detects deepfakes across all three modalities.

Deepfake Audio

AI generated speech designed to sound like a specific human speaker.

Deepfake audio refers to speech synthesised by machine learning models that replicate a target person's voice, tone, cadence, and accent. Modern systems can clone a voice from as little as three seconds of sample audio. Attackers use deepfake audio for CEO fraud, voice phishing, political disinformation, and unauthorized account access. Raid AI detects deepfake audio in real time with 99.7% accuracy across 120+ languages.

Deepfake Detection

Technology that identifies deepfakes and other AI generated media.

Deepfake detection is the process of analysing audio, image, or video content to determine whether it was generated or manipulated by AI rather than captured from a real source. Audio deepfake detection examines spectral patterns, vocal biomarkers, and temporal inconsistencies that distinguish genuine human speech from synthetic voices. Image and video deepfake detection analyses pixel level artifacts, facial geometry inconsistencies, temporal coherence across frames, and compression signatures left by generative models. Raid AI returns audio verdicts in under 50 milliseconds, fast enough to protect live calls and meetings, and checks images, video and documents in the same platform.

Deepfake Image

AI generated or manipulated face images used for impersonation or fraud.

A deepfake image is a photograph that has been generated or manipulated by AI to depict a person in a fabricated scenario. Techniques include face swaps that transplant one person's face onto another's body, fully synthetic faces generated from scratch by models like GANs and diffusion networks, and targeted tampering that alters specific facial features in authentic photographs. Deepfake images are used for identity fraud, nonconsensual content, forged documents, and disinformation. Raid AI detects deepfake images by analysing pixel level artifacts, facial geometry, lighting inconsistencies, and generative model signatures.

Deepfake Video

Face swapped, lip synced, or fully AI generated video content.

A deepfake video is video content that has been generated or manipulated by AI to alter the appearance, expressions, or actions of people in the footage. Common techniques include face swapping one person onto another's body across video frames, lip sync manipulation that makes a person appear to say words they never spoke, and fully synthetic video generated from text or image prompts. Deepfake videos are used for fraud, political disinformation, and nonconsensual content. Raid AI detects deepfake videos by analysing frame-level visual artifacts, temporal coherence, facial landmark consistency, and compression signatures unique to generative models.

Face Swap

Replacing one person's face with another in images or video using AI.

A face swap is an AI technique that replaces one person's face with another's in an image or video while attempting to preserve realistic lighting, skin tone, and facial expressions. Face swap models use encoder-decoder architectures or diffusion networks to map facial features between source and target identities. While face swaps have legitimate uses in film production and entertainment, they are widely misused for identity fraud, nonconsensual imagery, and impersonation scams. Raid AI detects face swaps by identifying blending boundary artifacts, inconsistent facial geometry, and unnatural texture transitions that betray the manipulation.

Generative Audio AI

AI systems that create new audio content from scratch.

Generative audio AI refers to machine learning models capable of producing new audio content, including speech, music, and sound effects. Modern generative audio models such as diffusion-based voice synthesisers and large audio language models can produce hyper-realistic speech in seconds. The rapid advance of generative audio AI is the core reason enterprises need real time deepfake detection like Raid AI.

Identity Theft

Unauthorized use of another person's identifying information or likeness.

Identity theft is the fraudulent acquisition and use of a person's private identifying information. In the AI era, identity theft has expanded beyond stolen passwords and credit card numbers to include cloned voices, synthetic face images, deepfake videos, and fabricated biometric signals. Audio based identity theft uses deepfake voice calls to impersonate victims, while image- and video based identity theft leverages face swaps and synthetic likenesses to bypass visual verification, forge documents, or create fraudulent content. Raid AI detects deepfakes across audio, image, and video to protect against the full spectrum of AI powered identity theft.

Multimodal Deepfake

A deepfake that combines multiple media types, such as synced video and audio.

A multimodal deepfake combines several deepfake formats, most commonly lip synced video with deepfake audio, to create a more convincing impersonation. Multimodal deepfakes are harder to debunk because viewers receive consistent signals across both channels. Detecting them requires specialized tooling for each modality. Raid AI detects deepfakes across all three modalities -- audio, image, and video -- providing comprehensive protection against multimodal attacks.

Spectral Analysis

Examination of the frequency components of an audio signal.

Spectral analysis is the process of decomposing an audio signal into its constituent frequencies to reveal patterns that are not visible in the raw waveform. Deepfake audio often contains subtle spectral artifacts, such as unnatural harmonic ratios, missing breathing noise, or abnormal spectral rolloff, that AI generators struggle to reproduce perfectly. Raid AI uses spectral analysis as one component of its multi-signal detection pipeline.

Synthetic Media

Any media created or significantly modified by artificial intelligence.

Synthetic media is a broad term for any audio, image, video, or text content produced or substantially altered by AI. Synthetic media includes deepfakes, AI generated art, machine translation, and auto generated news articles. While synthetic media has many legitimate uses, its misuse for fraud, disinformation, and harassment has created urgent demand for detection tools that can reliably distinguish synthetic content from authentic human-produced media.

Synthetic Voice

Machine generated audio that mimics human speech.

A synthetic voice is any audio output produced by a computer system to imitate human speech. Synthetic voices range from the obviously robotic outputs of early text to speech systems to modern AI generated voices that are indistinguishable from real humans. In a security context, the term synthetic voice usually refers to AI generated speech used with deceptive intent, which Raid AI is purpose built to detect.

Text to speech (TTS)

Technology that converts written text into spoken audio.

Text to speech (TTS) is a speech synthesis technology that converts written text into audible speech. Early TTS systems produced clearly robotic output, but neural TTS models now generate speech that is often indistinguishable from human recordings. When combined with voice cloning, TTS becomes the engine behind modern audio deepfakes, enabling attackers to make a cloned voice say anything they type.

Voice Biometrics

Authentication and identification based on unique vocal characteristics.

Voice biometrics is the use of a person's unique vocal characteristics, including pitch, cadence, and vocal tract shape, to authenticate or identify them. Voice biometric systems are widely used by banks and call centres for customer verification. However, modern voice cloning can spoof these systems, which is why organisations increasingly pair voice biometrics with deepfake detection to verify that the voice is both the right person AND a real human.

Voice Cloning

AI replication of a target person's voice from sample audio.

Voice cloning is the use of machine learning to recreate a specific person's voice from a short audio sample. Modern voice cloning systems can produce convincing imitations from just a few seconds of reference audio, then synthesise arbitrary speech in that voice. Legitimate uses include accessibility tools and content creation, but voice cloning is also weaponised for fraud, harassment, and disinformation campaigns.

Voice Conversion

Transforming one person's speech to sound like another person.

Voice conversion is a technique that takes a real recording of one speaker and transforms it to sound like a different target speaker, while preserving the original words and timing. Unlike voice cloning, which generates audio from text, voice conversion modifies existing audio in real time. This makes it particularly dangerous for live attacks during phone calls, where an attacker can speak naturally while the victim hears a trusted voice.

Voice Impersonation

Pretending to be a specific person using their voice or a realistic imitation.

Voice impersonation is the act of speaking in another person's voice with the intent of being mistaken for them. Historically, voice impersonation required talented human mimics. Today, AI voice cloning has democratised voice impersonation, allowing anyone with minimal technical skill to impersonate any public figure or target whose voice has been sampled online. Raid AI detects voice impersonation produced by both human and AI sources.

Voice Phishing (Vishing)

Phone based social engineering attacks that exploit voice communication.

Voice phishing, or vishing, is a social engineering attack that uses voice calls to trick victims into divulging sensitive information, authorising transactions, or granting access. Traditional vishing relies on spoofed caller ID and scripted impersonation. Modern vishing now incorporates deepfake audio to impersonate trusted executives, family members, or service representatives, making the attacks substantially harder to detect without AI powered deepfake detection.

Zero day Deepfake

A deepfake created with a voice generation model that has never been seen before.

A zero day deepfake is audio produced by a voice generation model that has not yet been catalogued or studied by detection systems. Just as zero day exploits bypass known-vulnerability databases, zero day deepfakes evade detection tools that rely solely on model fingerprinting. Robust deepfake detection, such as Raid AI's approach, uses generalisation-focused models that detect synthetic audio regardless of the source generator.