README.md

LFM2-Audio Arabic TTS

Loading

from liquid_audio import LFM2AudioModel, LFM2AudioProcessor
import torch

# Load base model and processor
processor = LFM2AudioProcessor.from_pretrained("LiquidAI/LFM2-Audio-1.5B")
model = LFM2AudioModel.from_pretrained("LiquidAI/LFM2-Audio-1.5B").eval().cuda()

# Load fine-tuned weights
state_dict = torch.load("pytorch_model.bin")
model.load_state_dict(state_dict)

Training: Stage 1 (Depthformer only), 10 epochs, Audio loss: 0.058