Scholarlink

With the growth in AI-generated material, a new issue has emerged in the form of deepfake music. These deepfakes are constructed with advanced models that can mimic human sounds or music in a variety of languages and accents. In this paper, we investigate the identification of multilingual music deepfakes - leveraging data from four languages: German, English, Spanish, and French. Regardless of the language or accent used, we aim to develop a system that can recognize phony music clips. We propose a novel framework that combines embeddings from WavLM-base-plus-sv, a speech and speaker-centric model, and MERT-v1-330M, an acoustic music understanding transformer, with a lightweight multilayer perceptron classifier to distinguish between real and deepfake music samples.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multilingual Music Deepfake Detection Using WavLM and MERT Audio Embeddings

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tejaswini Ponnada,
  • Pawani Agarwal,
  • Ignisha Rajathi G,
  • Yasir Abdullah R,
  • Mohanalin J,
  • Johny Elton R

摘要

With the growth in AI-generated material, a new issue has emerged in the form of deepfake music. These deepfakes are constructed with advanced models that can mimic human sounds or music in a variety of languages and accents. In this paper, we investigate the identification of multilingual music deepfakes - leveraging data from four languages: German, English, Spanish, and French. Regardless of the language or accent used, we aim to develop a system that can recognize phony music clips. We propose a novel framework that combines embeddings from WavLM-base-plus-sv, a speech and speaker-centric model, and MERT-v1-330M, an acoustic music understanding transformer, with a lightweight multilayer perceptron classifier to distinguish between real and deepfake music samples.