Fix build for current Spaces runtime

README.md

@@ -4,7 +4,7 @@ emoji: 💻
 colorFrom: yellow
 colorTo: gray
 sdk: gradio
-sdk_version: 5.38.0
+sdk_version: 6.30.0
 app_file: app.py
 pinned: false
 short_description: Demo for DMOSpeech 2

app.py

@@ -9,10 +9,27 @@ import numpy as np
 import spaces
 import torch
 import torchaudio
+import soundfile as sf
+import librosa
 from cached_path import cached_path
 from huggingface_hub import hf_hub_download
 from transformers import pipeline
 
+
+
+def _load_audio(path, *args, **kwargs):
+    # torchaudio>=2.9 routes load() through torchcodec; read files with soundfile/librosa instead
+    try:
+        data, sr = sf.read(path, dtype="float32", always_2d=True)
+        wav = torch.from_numpy(data.T.copy())
+    except Exception:
+        y, sr = librosa.load(path, sr=None, mono=False)
+        wav = torch.from_numpy(np.atleast_2d(y).astype(np.float32))
+    return wav, sr
+
+
+torchaudio.load = _load_audio
+
 from infer import DMOInference
 
 ## CUDA DEVICE ##

infer.py

@@ -112,7 +112,7 @@ class DMOInference:
         )
 
         # Load student checkpoint
-        checkpoint = torch.load(student_checkpoint_path, map_location="cpu")
+        checkpoint = torch.load(student_checkpoint_path, map_location="cpu", weights_only=False)
         self.model.load_state_dict(checkpoint["model_state_dict"], strict=False)
 
         # Setup generator and teacher
@@ -158,7 +158,7 @@ class DMOInference:
 
         self.SLP.eval()
         self.SLP.load_state_dict(
-            torch.load(checkpoint_path, map_location="cpu")["model_state_dict"]
+            torch.load(checkpoint_path, map_location="cpu", weights_only=False)["model_state_dict"]
         )
 
     def predict_duration(

requirements.txt

@@ -1,41 +1,40 @@
-# Main dependencies
-accelerate>=0.33.0
-# Note: bitsandbytes has platform restrictions - not for arm64 or Darwin (macOS)
-bitsandbytes>0.37.0
-cached_path
-click
-datasets
-ema_pytorch>=0.5.2
-gradio>=3.45.2
-hydra-core>=1.3.0
-jieba
-librosa
-matplotlib
-numpy<=1.26.4
-pydantic<=2.10.6
-pydub
-pypinyin
-safetensors
-soundfile
-tomli
-torch>=2.0.0
-torchaudio>=2.0.0
-torchdiffeq
-tqdm>=4.65.0
-transformers
-transformers_stream_generator
-unidecode
-vocos
-wandb
-x_transformers>=1.31.14
-
-# Optional dependencies for evaluation
-# Uncomment if you need eval features:
-# faster_whisper==0.10.1
-# funasr
-# jiwer
-# modelscope
-# zhconv
-# zhon
-
+# Main dependencies
+accelerate>=0.33.0
+# Note: bitsandbytes has platform restrictions - not for arm64 or Darwin (macOS)
+bitsandbytes>0.37.0
+cached_path
+click
+datasets
+ema_pytorch>=0.5.2
+gradio>=3.45.2
+hydra-core>=1.3.0
+jieba
+librosa
+matplotlib
+numpy<=1.26.4
+pydub
+pypinyin
+safetensors
+soundfile
+tomli
+torch==2.11.0
+torchaudio==2.11.0
+torchdiffeq
+tqdm>=4.65.0
+transformers>=5.0
+transformers_stream_generator
+unidecode
+vocos
+wandb
+x_transformers>=1.31.14
+
+# Optional dependencies for evaluation
+# Uncomment if you need eval features:
+# faster_whisper==0.10.1
+# funasr
+# jiwer
+# modelscope
+# zhconv
+# zhon
+
 cached_path