Upload 3 files

app.py

@@ -0,0 +1,102 @@
+from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+import torch
+from torchvision import models, transforms
+import torch.nn.functional as F
+import librosa, soundfile as sf, tempfile
+import numpy as np
+import matplotlib.pyplot as plt
+import librosa.display
+from PIL import Image
+import io
+from feature_extract import AudioFeatureExtractor
+import requests, os
+
+# === CONFIG ===
+MODEL_REPO = "Chula-PD/voice-mobilenet-pd"
+MODEL_FILE = "MobileNet_Model.pth"
+MODEL_URL = f"https://huggingface.co/{MODEL_REPO}/resolve/main/{MODEL_FILE}"
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# === FastAPI Init ===
+app = FastAPI(title="CheckPD Voice API", version="1.0")
+
+# Allow CORS (for React frontend)
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],  # ปรับให้เฉพาะ domain ได้ภายหลัง
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# === Load Model ===
+def load_model():
+    if not os.path.exists(MODEL_FILE):
+        print("Downloading model weights from Hugging Face...")
+        weights_bytes = requests.get(MODEL_URL)
+        with open(MODEL_FILE, "wb") as f:
+            f.write(weights_bytes.content)
+    model = models.mobilenet_v3_small(weights=None)
+    in_features = model.classifier[-1].in_features
+    model.classifier[-1] = torch.nn.Linear(in_features, 2)
+    model.load_state_dict(torch.load(MODEL_FILE, map_location=device))
+    model.eval()
+    return model
+
+model = load_model()
+classes = ["HC", "PD"]
+
+# === Image Transform ===
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor(),
+    transforms.Normalize(
+        [0.485, 0.456, 0.406],
+        [0.229, 0.224, 0.225]
+    ),
+])
+
+@app.get("/")
+def home():
+    return {"message": "CheckPD Voice API is running."}
+
+@app.post("/predict")
+async def predict(file: UploadFile = File(...)):
+    try:
+        # Load and preprocess audio
+        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
+            tmp.write(await file.read())
+            tmp.flush()
+            wav_path = tmp.name
+
+        extractor = AudioFeatureExtractor(wav_path, sr=16000)
+        mel_db = extractor.get_melspectrogram()
+
+        # Convert mel to image
+        fig, ax = plt.subplots(figsize=(6, 3))
+        librosa.display.specshow(mel_db, sr=16000, hop_length=51, cmap="viridis")
+        plt.axis("off")
+        buf = io.BytesIO()
+        plt.savefig(buf, format='png', bbox_inches="tight", pad_inches=0)
+        plt.close()
+        buf.seek(0)
+        image = Image.open(buf).convert("RGB")
+
+        # Predict
+        input_tensor = transform(image).unsqueeze(0).to(device)
+        with torch.no_grad():
+            outputs = model(input_tensor)
+            probs = F.softmax(outputs, dim=1)
+            pred_idx = torch.argmax(probs, dim=1).item()
+            confidence = probs[0][pred_idx].item()
+
+        return {
+            "label": classes[pred_idx],
+            "confidence": round(confidence, 4)
+        }
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))

feature_extract.py

@@ -0,0 +1,82 @@
+import torch
+import numpy as np
+import librosa
+import librosa.display
+import matplotlib.pyplot as plt
+
+
+class AudioFeatureExtractor:
+    def __init__(self, wavfile, sr=16000, n_fft=1024, hop_length=51, n_mels=256):
+        self.wavfile = wavfile
+        self.target_sr = sr
+        self.n_fft = n_fft
+        self.hop_length = hop_length
+        self.n_mels = n_mels
+
+        # ✅ โหลดเสียงด้วย librosa (resample อัตโนมัติ)
+        waveform, _ = librosa.load(self.wavfile, sr=self.target_sr)
+        waveform = torch.tensor(waveform).unsqueeze(0)
+        self.waveform = waveform
+        self.sr = self.target_sr
+
+    def get_spectrogram(self, to_db=True):
+        """สร้าง spectrogram แบบธรรมดา"""
+        spec = np.abs(librosa.stft(
+            self.waveform.squeeze(0).numpy(),
+            n_fft=self.n_fft,
+            hop_length=self.hop_length
+        )) ** 2
+        if to_db:
+            spec = librosa.power_to_db(spec, ref=np.max)
+        return spec
+
+    def get_melspectrogram(self):
+        """สร้าง Mel-spectrogram"""
+        mel_spec = librosa.feature.melspectrogram(
+            y=self.waveform.squeeze(0).numpy(),
+            sr=self.sr,
+            n_fft=self.n_fft,
+            hop_length=self.hop_length,
+            n_mels=self.n_mels,
+            power=2.0
+        )
+        mel_db = librosa.power_to_db(mel_spec, ref=np.max)
+        return mel_db
+
+    def normalize(self, spec):
+        """ปรับค่าสีให้อยู่ในช่วง 0–1"""
+        spec_min, spec_max = spec.min(), spec.max()
+        return (spec - spec_min) / (spec_max - spec_min + 1e-6)
+
+    def to_grayscale(self, spec):
+        """แปลงให้เป็น 1-channel"""
+        return np.expand_dims(spec, axis=0)
+
+    def get_normalized_melspec(self):
+        mel_db = self.get_melspectrogram()
+        mel_norm = self.normalize(mel_db)
+        return self.to_grayscale(mel_norm)
+
+    def plot_melspectrogram(self, save_path=None):
+        mel_db = self.get_melspectrogram()
+        plt.figure(figsize=(10, 4))
+        librosa.display.specshow(mel_db, sr=self.sr, hop_length=self.hop_length, cmap="viridis")
+        plt.axis("off")
+        plt.tight_layout()
+        if save_path:
+            plt.savefig(save_path, bbox_inches="tight", pad_inches=0)
+            plt.close()
+        else:
+            plt.show()
+
+    def save_melspectrogram(self, out_path="melspec.png"):
+        melspec = self.get_melspectrogram()
+        plt.figure(figsize=(10, 4))
+        import librosa.display
+
+        librosa.display.specshow(melspec, sr=self.sr, hop_length=self.hop_length)
+        plt.axis("off")
+        plt.tight_layout()
+        plt.savefig(out_path, bbox_inches="tight", pad_inches=0)
+        plt.close()
+        return out_path

requirements.txt

@@ -0,0 +1,10 @@
+fastapi
+uvicorn
+torch
+torchvision
+librosa
+soundfile
+matplotlib
+pillow
+numpy
+requests