Detection of AI-Generated Audio: Speech, Environmental Sound, Music, and Beyond

Audio deepfakes, by definition, are synthetic audio recordings generated using deep learning-based systems for either malicious, artistic, or entertainment purposes. In most cases, an audio deepfake can be compared to an authentic, natural audio recording. For example, in the detection of spoofed speech, samples of real and falsified speech are compared. More generally speaking, AI generated audio content has rapidly evolved with recent advances in generative modeling, ranging from high-fidelity text-to-speech and voice conversion systems to sophisticated audio diffusion models capable of producing realistic sound events, ambient environments, and music. It is therefore important to attempt to provide an in-depth overview of recent advances in the detection of AI-generated audio, by compiling a series of articles on this subject.The goal of this Research Topic is to explore, formalize, and advance detection frameworks that can operate either on single modalities or robustly across diverse forms of synthetic audio—combining speech, paralinguistic vocalizations, sound events, audio scenes, and music. We are also interested in systems handling AI-generated content beyond audio, namely multimodal deepfakes, such as audio-video synthetic content. Current systems are typically domain-specific and struggle to generalize beyond the classes of signals they were trained on. As generative audio models continue to converge in architecture and quality, there is a need for detectors capable of reasoning across signal categories, frequency content, and temporal structures.We seek contributions that investigate models, representations, datasets, and evaluation methodologies enabling broad-spectrum AI-generated audio content detection. Potential directions include, but are not limited to: generalizable feature learning, self-supervised or foundation-model-based representations, domain-invariant training, cross-modal architectures, hybrid anomaly-detection pipelines, and methods that explicitly capture the shared artefacts of generative processes across speech and non-speech domains. We also welcome work addressing robustness to compression, environmental noise, adversarial attacks, partial forgeries, or mixed-source synthetic content. Ultimately, this Research Topic seeks to foster the development of next-generation systems that can withstand the growing diversity of synthetic audio and support trustworthy real-world deployment.We welcome original contributions addressing any aspect related to the detection of AI-generated audio and audio deepfakes, including, but not limited to, studies on:● detection frameworks for speech, music, everyday sounds, non-speech synthetic audio and multimodal media such as audio-video;● foundation models or self-supervised audio representations for general-purpose detection;● complementary approaches such as audio and speech watermarking;● cross-domain and cross-dataset evaluation protocols;● domain adaptation and generalization to unseen attacks and datasets;● detection of mixed or partially forged audio;● adversarial attacks and defenses in audio deepfake detection;● unified or universal models for audio deepfake detection;● forensic analysis, interpretability, explainability and artefact characterization;● robustness to real-world conditions (noise, codecs, recording devices);● ethical, societal, and legal considerations of audio deepfake detection;● real-time, streaming, and low-latency audio deepfake detection systems;● datasets including speech and/or non-speech deepfakes/genAI content;● metrics and benchmarks for broad-coverage detection.Topic Editors Dr. Rohan Kumar Das and Dr. Elie Khoury are employed by Fortemedia and Pindrop, respectively. All other Topic Editors declare no competing interests with regards to the Research Topic subject. Topic editors: Thomas Pellegrini, Elie Khoury, PhD, Emmanouil Benetos, Sayaka Shiota, Rohan Kumar Das Source: https://www.frontiersin.org/research-topics/77553/detection-of-ai-generated-audio-speech-environmental-sound-music-and-beyond