chore: tidy Space repository structure

README.md

@@ -20,11 +20,6 @@ tags:
   - music
   - hackathon
   - build-small
-  - track:wood
-  - sponsor:openbmb
-  - sponsor:openai
-  - achievement:offgrid
-  - achievement:fieldnotes
 short_description: AI-powered kinetic typography lyric video generator
 ---
 
@@ -150,6 +145,3 @@ https://youtu.be/6RJwgFu6LHQ
 
 ## Tested on:
 RTX 5060 ti 16 GB
-
-## SOCIAL MEDIA POST:
-https://dev.to/blazestorm/amuseme-when-small-models-compose-a-visual-symphony-50fc

packages.txt

@@ -0,0 +1,2 @@
+fonts-dejavu-core
+fonts-liberation

requirements.txt

@@ -3,10 +3,11 @@
 # torch/torchaudio come from the ZeroGPU runtime + the demucs dependency,
 # so they are intentionally not pinned here.
 
+# ctranslate2: do NOT pin — let pip resolve the wheel matching ZeroGPU's
+# CUDA/cuDNN version. Pinning 4.3.1 (cuDNN 8 build) breaks on ZeroGPU's cuDNN 9
+# with "libcudnn_ops_infer.so.8: cannot open shared object file".
 faster-whisper
-ctranslate2==4.3.1
 demucs
-torchcodec
 pillow
 pydantic
 spaces

src/amuseme/transcriber.py

@@ -70,15 +70,24 @@ def _load_model(model_size: str = "large-v3"):
     if _model is None:
         if model_size == "turbo":
             model_size = "large-v3-turbo"
-            
+
         device = "cpu" if os.environ.get("FORCE_CPU") == "1" else "cuda"
         logger.info(f"Loading Whisper {model_size} on {device}...")
         compute_type = "float16" if device == "cuda" else "int8"
         try:
             _model = WhisperModel(model_size, device=device, compute_type=compute_type)
         except Exception as e:
-            logger.warning(f"Failed to load {model_size} with {compute_type}: {e}. Falling back to float32.")
-            _model = WhisperModel(model_size, device=device, compute_type="float32")
+            if device == "cuda":
+                # CUDA/cuDNN library mismatch (e.g. on ZeroGPU). Retry on CPU.
+                logger.warning(f"CUDA load failed ({e}). Falling back to CPU int8.")
+                try:
+                    _model = WhisperModel(model_size, device="cpu", compute_type="int8")
+                except Exception as e2:
+                    logger.error(f"CPU fallback also failed: {e2}")
+                    raise
+            else:
+                logger.warning(f"Failed to load with {compute_type}: {e}. Retrying with float32.")
+                _model = WhisperModel(model_size, device=device, compute_type="float32")
     return _model