Data Analytics Lab Announces Research Results on Generative AI Voice Determination Technology
Data Analytics Lab Co., Ltd. announced research results on AI-generated voice detection technology, conducted in collaboration with Evixar Inc. for a Ministry of Internal Affairs and Communications project. The research aims to advance deepfake countermeasures.
📋 Article Processing Timeline
- 📰 Published: April 9, 2026 at 21:00
- 🔍 Collected: April 9, 2026 at 12:30
- 🤖 AI Analyzed: April 20, 2026 at 08:35 (260h 5m after Collected)
Data Analytics Lab Co., Ltd. (Headquarters: Chuo-ku, Tokyo, CEO: Masahiko Kondo, hereinafter "our company") announced the compilation of research results on technology for determining audio content generated by AI. This was conducted as part of a joint research project where Evixar Inc. (hereinafter "Evixar") was selected for the Ministry of Internal Affairs and Communications' "Development and Demonstration Project for Countermeasure Technologies against Fake/Misinformation on the Internet."
This research integrates Evixar's acoustic signal processing technology and our company's AI/data analysis technology, aiming to advance countermeasures against fake and misinformation (deepfakes, etc.) that are becoming more serious with the sophistication of generative AI.
■ Overview of Research Results
In this research, we reproduced and analyzed the characteristics of audio content by generative AI, and built verification data and an AI model verification environment for determining synthesized audio.
■ Main Results
① Construction of a verification platform supporting diverse audio generation models
Towards the analysis of synthesized audio including Japanese, we investigated and compared advanced audio generation models such as:
・Tortoise
・XTTS (multilingual model)
・Qwen3-TTS
and conducted verifications on multiple generation methods.
In particular, by supporting audio generation technologies based on multilingual and large-scale learning like XTTS, we conducted verification under conditions close to real generative AI environments.
② Systematic generation and feature analysis of synthesized audio data
In this research, we conducted:
・Organization and systematization of generation conditions for synthesized audio data
・Analysis of audio signals (spectrograms, etc.)
・Extraction of structural differences from natural audio
and worked on quantitatively grasping the characteristics of synthesized audio.
Through this, we worked towards the research and development of a versatile determination technology independent of generation models.
③ Verification of synthesized audio determination models using deep learning
Regarding the determination of generated audio, we:
・Investigated and verified deep learning models
・Built learning datasets
・Prepared evaluation processes for determination accuracy
and confirmed the effectiveness of determination models utilizing features unique to AI audio to a certain extent.
④ Technological advancement through the fusion of acoustic signal processing and AI
In this research, we combined Evixar's acoustic signal processing technology with our AI technology to support the verification for strengthening Evixar's synthesized audio determination system (EAF). Specifically, we:
・Generated synthesized audio data and verified the operation of diverse generation models
・Grasped the differences between synthesized and natural audio through feature analysis of audio signals
・Verified determination accuracy using deep learning models and built evaluation datasets
and worked on providing technical insights to improve EAF's determination accuracy.
■ Positioning of this Research
This research was promoted as a project adopted by the Ministry of Internal Affairs and Communications led by Evixar Inc., where they are responsible for developing acoustic signal processing and AI countermeasure technologies, and our company is responsible for the data design, analysis, and verification areas.
Our company specifically took charge of:
・Generation and design of synthesized audio data
・Analysis and feature extraction of audio data
・Verification of determination models and construction of evaluation platforms
and in this...
This research integrates Evixar's acoustic signal processing technology and our company's AI/data analysis technology, aiming to advance countermeasures against fake and misinformation (deepfakes, etc.) that are becoming more serious with the sophistication of generative AI.
■ Overview of Research Results
In this research, we reproduced and analyzed the characteristics of audio content by generative AI, and built verification data and an AI model verification environment for determining synthesized audio.
■ Main Results
① Construction of a verification platform supporting diverse audio generation models
Towards the analysis of synthesized audio including Japanese, we investigated and compared advanced audio generation models such as:
・Tortoise
・XTTS (multilingual model)
・Qwen3-TTS
and conducted verifications on multiple generation methods.
In particular, by supporting audio generation technologies based on multilingual and large-scale learning like XTTS, we conducted verification under conditions close to real generative AI environments.
② Systematic generation and feature analysis of synthesized audio data
In this research, we conducted:
・Organization and systematization of generation conditions for synthesized audio data
・Analysis of audio signals (spectrograms, etc.)
・Extraction of structural differences from natural audio
and worked on quantitatively grasping the characteristics of synthesized audio.
Through this, we worked towards the research and development of a versatile determination technology independent of generation models.
③ Verification of synthesized audio determination models using deep learning
Regarding the determination of generated audio, we:
・Investigated and verified deep learning models
・Built learning datasets
・Prepared evaluation processes for determination accuracy
and confirmed the effectiveness of determination models utilizing features unique to AI audio to a certain extent.
④ Technological advancement through the fusion of acoustic signal processing and AI
In this research, we combined Evixar's acoustic signal processing technology with our AI technology to support the verification for strengthening Evixar's synthesized audio determination system (EAF). Specifically, we:
・Generated synthesized audio data and verified the operation of diverse generation models
・Grasped the differences between synthesized and natural audio through feature analysis of audio signals
・Verified determination accuracy using deep learning models and built evaluation datasets
and worked on providing technical insights to improve EAF's determination accuracy.
■ Positioning of this Research
This research was promoted as a project adopted by the Ministry of Internal Affairs and Communications led by Evixar Inc., where they are responsible for developing acoustic signal processing and AI countermeasure technologies, and our company is responsible for the data design, analysis, and verification areas.
Our company specifically took charge of:
・Generation and design of synthesized audio data
・Analysis and feature extraction of audio data
・Verification of determination models and construction of evaluation platforms
and in this...