Zifeng Radio Tone RVC Model (48k Self-Training, Commercial Available)

Zifeng Radio broadcast timbre, based on Applio's self-trained RVC v2 model, with a sampling rate of 48k.

The dataset consisted of 145 clean dry sounds (approximately 15 minutes 44 seconds in total, monolophras), which were processed by contentvec feature extraction + rmvpe pitch extraction, and trained with 200 epochs / 21,000 steps using Applio HiFi-GAN pretrained base molds (f0G48k/f0D48k).

Features: Restores the relaxed feel and articulation of radio broadcasting, suitable for reading, broadcasting, narration, and oral content.

Usage recommendation: Inference sampling rate should be 48000.