A practical method for generating whispers from singing voices: Application of improved phantom silhouette method

The previously proposed phantom silhouette method is promising for converting ordinary speech into whispered speech. It a simple parametric that uses high-quality vocoder-type analysis and synthesis. An sample first analyzed using the WORLD vocoder. Then, based on extracted spectral envelope, features are manipulated so voice sounds like whisper. target synthesized by driving it with white noise instead of vocal source signal to make whole sound voiceless. In this study, was applied singing voices generate whisper voices. addition actual voices, virtual singers' were generated Vocaloid synthesizer, AI NEUTRINO neural synthesizer also tested from