Create app.py

app.py

@@ -0,0 +1,58 @@
+import os
+import torch
+import gradio as gr
+from transformers import pipeline
+
+# Model configuration
+MODEL_NAME = "chuuhtetnaing/whisper-large-v3-myanmar"
+
+# Automatically use GPU if available on HF Space, fallback to CPU
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+
+print(f"Loading pipeline on {device}...")
+
+# Load model pipeline ONCE when the Space starts up
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model=MODEL_NAME,
+    torch_dtype=torch_dtype,
+    device=device,
+    chunk_length_s=30,  # Splice long audio into 30s chunks automatically
+)
+
+def transcribe_audio(audio_file):
+    if audio_file is None:
+        return "Please upload or record an audio snippet."
+
+    # Run transcription
+    outputs = pipe(
+        audio_file,
+        generate_kwargs={
+            "language": "burmese", 
+            "task": "transcribe"
+        },
+        return_timestamps=True,
+    )
+    
+    return outputs["text"]
+
+# Define Gradio Interface
+demo = gr.Interface(
+    fn=transcribe_audio,
+    inputs=gr.Audio(
+        type="filepath", 
+        label="Record or Upload Audio",
+        sources=["microphone", "upload"]
+    ),
+    outputs=gr.Textbox(
+        label="Myanmar Transcription Result", 
+        lines=6, 
+        show_copy_button=True
+    ),
+    title="🇲🇲 Myanmar Speech-to-Text (Whisper Large v3)",
+    description="Upload an audio file (`.wav`, `.mp3`, `.m4a`) or record directly from your microphone to get Myanmar transcriptions.",
+)
+
+if __name__ == "__main__":
+    demo.launch()