Create app.py

app.py

@@ -0,0 +1,85 @@
+from fastapi import FastAPI, UploadFile, File, HTTPException
+import uvicorn
+import torchaudio
+import torch.nn.functional as F
+import torch
+import numpy as np
+import onnxruntime as ort
+from huggingface_hub import hf_hub_download
+import os
+
+app = FastAPI(title="Pakistani LID AI Engine (Standalone)")
+
+print("📥 Checking/Downloading ONNX Model from Hugging Face...")
+# Yeh line model ko cache kar legi, baar baar download nahi karegi
+model_path = hf_hub_download(repo_id="Hammad712/pakistani-lid-v3-sota", filename="pakistani_lid_v3.onnx")
+
+print("🚀 Loading ONNX Session for CPU...")
+session = ort.InferenceSession(model_path, providers=['CPUExecutionProvider'])
+
+labels = ("balochi", "english", "pashto", "sindhi", "urdu")
+id2label = {i: label for i, label in enumerate(labels)}
+
+def predict_audio(audio_path):
+    waveform, sr = torchaudio.load(audio_path)
+    if waveform.shape[0] > 1: waveform = waveform.mean(dim=0, keepdim=True)
+    if waveform.ndim == 1: waveform = waveform.unsqueeze(0)
+    
+    target_frames = int(sr * 15)
+    if waveform.shape[1] > target_frames: waveform = waveform[:, :target_frames]
+    if sr != 16000: waveform = torchaudio.functional.resample(waveform, sr, 16000)
+    
+    peak = waveform.abs().max().clamp(min=1e-6)
+    waveform = (waveform / peak) - waveform.mean()
+    waveform = waveform / waveform.std().clamp(min=1e-6)
+    
+    length = waveform.shape[1]
+    mask = torch.zeros(16000 * 15, dtype=torch.long)
+    if length >= 16000 * 15:
+        waveform, mask[:] = waveform[:, :16000 * 15], 1
+    else:
+        mask[:length] = 1
+        waveform = F.pad(waveform, (0, 16000 * 15 - length))
+
+    ort_inputs = {
+        "input_values": waveform.numpy(),
+        "attention_mask": mask.unsqueeze(0).numpy()
+    }
+    
+    logits = session.run(None, ort_inputs)[0]
+    exp_logits = np.exp(logits - np.max(logits, axis=1, keepdims=True))
+    probs = exp_logits / np.sum(exp_logits, axis=1, keepdims=True)
+    
+    pred_id = np.argmax(probs, axis=1)[0]
+    return id2label[pred_id], float(probs[0][pred_id])
+
+@app.post("/predict")
+async def predict_language(file: UploadFile = File(...)):
+    if not file.filename.endswith(('.wav', '.mp3', '.m4a', '.ogg')):
+        raise HTTPException(status_code=400, detail="Invalid audio format. Please upload wav, mp3, m4a, or ogg.")
+
+    temp_audio_path = f"temp_{file.filename}"
+    try:
+        # File save karein
+        with open(temp_audio_path, "wb") as buffer:
+            buffer.write(await file.read())
+
+        # Prediction lein
+        lang, confidence = predict_audio(temp_audio_path)
+        
+        # Temp file delete kar dein
+        os.remove(temp_audio_path)
+
+        return {
+            "success": True,
+            "language": lang.upper(),
+            "confidence": round(confidence * 100, 2)
+        }
+    except Exception as e:
+        if os.path.exists(temp_audio_path):
+            os.remove(temp_audio_path)
+        raise HTTPException(status_code=500, detail=str(e))
+
+if __name__ == "__main__":
+    print("✨ Server is LIVE at: http://localhost:8080")
+    uvicorn.run(app, host="0.0.0.0", port=8080)