init

app.py

@@ -5,7 +5,6 @@ import tempfile
 import gradio as gr
 import librosa
 import soundfile
-import tomli
 import torch
 import torch.nn.functional as F
 import torchaudio

src/moviedubber/infer_with_mmlm_result.py

@@ -71,7 +71,9 @@ def load_models(device):
 
     vocoder = load_vocoder(local_path="nvidia/bigvgan_v2_24khz_100band_256x", device=device)
 
-    model_path = hf_hub_download(repo_id="woak-oa/DeepDubber-V1")
+    ckpt_path = hf_hub_download(repo_id="woak-oa/DeepDubber-V1", filename="mmdubber.pt")
+    vocab_file = hf_hub_download(repo_id="woak-oa/DeepDubber-V1", filename="vocab.txt")
+    campplus_path = hf_hub_download(repo_id="woak-oa/DeepDubber-V1", filename="campplus.onnx")
 
     model_cls = DiT
     model_cfg = OmegaConf.load(model_cfg).model.arch
@@ -80,9 +82,9 @@ def load_models(device):
     ema_model = load_model(
         model_cls,
         model_cfg,
-        ckpt_path=f"{model_path}/mmdubber.pt",
+        ckpt_path=ckpt_path,
         mel_spec_type=vocoder_name,
-        vocab_file=f"{model_path}/vocab.txt",
+        vocab_file=vocab_file,
         controlnet=controlnet,
         device=device,
     )
@@ -92,7 +94,7 @@ def load_models(device):
     option.intra_op_num_threads = 1
     providers = ["CPUExecutionProvider"]
     ort_session = onnxruntime.InferenceSession(
-        f"{model_path}/campplus.onnx",
+        campplus_path,
         sess_options=option,
         providers=providers,
     )