EZ-Ar-VC
An Arabic-oriented voice conversion model for speech anonymization, built on the EZ-VC architecture and trained on diverse Arabic speech data. It preserves phonological features such as emphatic consonants and long vowels and is robust to code-switching and cross-dialect variation.
Timeline (last 90 days)
EZ-Ar-VC is an Arabic-centric voice anonymization model built on the EZ-VC architecture and trained on diverse Arabic speech data.
EZ-Ar-VC shows strong performance in intelligibility, naturalness and pronunciation preservation while maintaining high speaker distinctiveness.
Subjective evaluations show that EZ-Ar-VC-converted speech is close to the original speech in naturalness and has significantly fewer pronunciation errors.
EZ-Ar-VC can more accurately preserve Arabic-specific pronunciation details, such as emphatic consonants and long vowels.
The study verified through subjective evaluation the robustness of EZ-Ar-VC to code-switching and cross-dialect conversion, and the model can handle dialects with limited representation in the training data.
All Facts (5)
EZ-Ar-VC is an Arabic-centric voice anonymization model built on the EZ-VC architecture and trained on diverse Arabic speech data.
50%UnverifiedEZ-Ar-VC shows strong performance in intelligibility, naturalness and pronunciation preservation while maintaining high speaker distinctiveness.
50%UnverifiedSubjective evaluations show that EZ-Ar-VC-converted speech is close to the original speech in naturalness and has significantly fewer pronunciation errors.
50%UnverifiedEZ-Ar-VC can more accurately preserve Arabic-specific pronunciation details, such as emphatic consonants and long vowels.
50%UnverifiedThe study verified through subjective evaluation the robustness of EZ-Ar-VC to code-switching and cross-dialect conversion, and the model can handle dialects with limited representation in the training data.
50%