dare43321/snac-child-finetuned

SNAC Fine-tuned — Child Voice (Hindi/English)

Fine-tuned from hubertsiuzdak/snac_24khz on child voice audio data (Hindi + English).

Training

• Base model : hubertsiuzdak/snac_24khz
• Dataset : dare43321/mimi-snac-ready
• Epochs : 50
• Final loss : 0.1627
• Strategy : Encoder frozen, decoder + VQ fine-tuned

Usage

import torch
from snac import SNAC
from huggingface_hub import hf_hub_download

model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
weights = hf_hub_download("dare43321/snac-child-finetuned", "pytorch_model.bin")
model.load_state_dict(torch.load(weights, map_location="cpu"))
model.eval()