Would you release the flow matching & Hifi-GAN model weights?
#1
by
JosephusCheung
- opened
For role-aware speech synthesis, we utilize the pre-trained conditional flow matching and HiFi-GAN weights from GLM-4-Voice (Zeng et al., 2024) and fine-tune them on high-quality role speech data.
Sorry, due to company policy, we are unable to release the weights of the flow matching and HiFi-GAN models. We’ll be uploading character audio data later, and you can fine-tune using the pretrained GLM-4-Voice weights from that.
haonanzhang
changed discussion status to
closed