fixed AttributeError: 'int' object has no attribute 'ndim'

app.py

@@ -5,27 +5,36 @@ from transformers import WhisperProcessor, WhisperForConditionalGeneration
 
 MODEL_ID = "codeninja3d/whisper-small-ls100"
 
+# Load model + processor
 processor = WhisperProcessor.from_pretrained(MODEL_ID)
 model = WhisperForConditionalGeneration.from_pretrained(MODEL_ID)
 device = "cuda" if torch.cuda.is_available() else "cpu"
 model = model.to(device)
 
 def transcribe(audio):
-    if audio is None:
+    # Handle no input or bad input
+    if not audio or not isinstance(audio, tuple):
         return "⚠️ Please record or upload an audio clip."
 
-    # Gradio gives (np.ndarray, sample_rate)
     speech_array, sampling_rate = audio
 
-    # Ensure correct format
-    if isinstance(speech_array, tuple):
-        speech_array = np.array(speech_array[0], dtype=np.float32)
+    # Ensure it's a NumPy array of floats
+    if not isinstance(speech_array, np.ndarray):
+        try:
+            speech_array = np.array(speech_array, dtype=np.float32)
+        except Exception:
+            return "⚠️ Invalid audio input received."
 
-    # Whisper expects float32 mono audio
-    speech_array = np.mean(speech_array, axis=-1) if speech_array.ndim > 1 else speech_array
-    inputs = processor(speech_array, sampling_rate=sampling_rate, return_tensors="pt").to(device)
+    # Convert to mono if stereo
+    if speech_array.ndim > 1:
+        speech_array = np.mean(speech_array, axis=-1)
+
+    # Normalize range to [-1, 1]
+    if np.abs(speech_array).max() > 1.0:
+        speech_array = speech_array / np.abs(speech_array).max()
 
-    # Generate transcription
+    # Run Whisper
+    inputs = processor(speech_array, sampling_rate=sampling_rate, return_tensors="pt").to(device)
     pred_ids = model.generate(**inputs)
     text = processor.batch_decode(pred_ids, skip_special_tokens=True)[0]
     return text.strip()
@@ -39,4 +48,5 @@ demo = gr.Interface(
 )
 
 if __name__ == "__main__":
-    demo.launch(share=True)
+    # On Hugging Face Spaces, no need for share=True
+    demo.launch()

requirements.txt

@@ -2,3 +2,5 @@ torch
 torchaudio
 transformers>=4.45
 gradio>=4.0
+safetensors
+numpy