updated the app

app.py

@@ -1,6 +1,5 @@
 import gradio as gr
 import torch
-import torchaudio
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
 
 MODEL_ID = "codeninja3d/whisper-small-ls100"
@@ -10,19 +9,22 @@ model = WhisperForConditionalGeneration.from_pretrained(MODEL_ID)
 model = model.to("cuda" if torch.cuda.is_available() else "cpu")
 
 def transcribe(audio):
-    speech, sr = torchaudio.load(audio)
-    inputs = processor(speech.squeeze(), sampling_rate=sr, return_tensors="pt").to(model.device)
+    # audio is (numpy array, sample_rate)
+    if audio is None:
+        return "⚠️ No audio input detected."
+    speech, sr = audio
+    inputs = processor(speech, sampling_rate=sr, return_tensors="pt").to(model.device)
     pred_ids = model.generate(**inputs)
     text = processor.batch_decode(pred_ids, skip_special_tokens=True)[0]
     return text
 
 demo = gr.Interface(
     fn=transcribe,
-    inputs=gr.Audio(sources=["upload", "microphone"], type="filepath"),
+    inputs=gr.Audio(sources=["upload", "microphone"], type="numpy"),
     outputs="text",
-    title="Whisper Small LS100 - Fine-tuned Model",
-    description="Upload or record an English audio sample (16 kHz recommended) to get a transcription."
+    title="Whisper Small LS100 - Fine-tuned English ASR",
+    description="🎙 Upload or record audio (16 kHz) to get transcription."
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)