Upload app.py

app.py

@@ -0,0 +1,307 @@
+# (Paste your full Gradio app.py code here)
+!pip install -q gradio soundfile
+
+import gradio as gr
+from pathlib import Path
+
+# Use your project brief path from the notebook
+PROJECT_BRIEF_LOCAL = "/mnt/data/Project_description.pdf"
+
+
+def normalize_audio_path(audio_file):
+    """Return a filesystem path (or None) from Gradio's audio input."""
+    if audio_file is None:
+        return None
+    if isinstance(audio_file, str):
+        return audio_file
+    if hasattr(audio_file, "name"):
+        return audio_file.name
+    if isinstance(audio_file, (tuple, list)) and len(audio_file) > 0:
+        # In some cases Gradio gives (filepath, ...other stuff)
+        return audio_file[0]
+    return str(audio_file)
+
+
+# Extended cuisine map: starters / main / desserts, separated veg / non-veg
+CUISINE_MAP = {
+    "kerala": {
+        "region": "Kerala",
+        "starters_veg": ["Banana Chips", "Vegetable Uzhunnu Vada"],
+        "starters_nonveg": ["Kerala Fish Fry", "Chicken 65 (Kerala Style)"],
+        "main_veg": ["Appam with Vegetable Stew", "Puttu with Kadala Curry"],
+        "main_nonveg": ["Kerala Fish Curry", "Beef Ularthiyathu", "Chicken Stew"],
+        "desserts": ["Palada Payasam", "Ada Pradhaman"],
+    },
+
+    "hindi": {
+        "region": "North India (Hindi-speaking belt)",
+        "starters_veg": ["Paneer Tikka", "Veg Samosa"],
+        "starters_nonveg": ["Chicken 65", "Mutton Seekh Kabab"],
+        "main_veg": ["Paneer Butter Masala", "Dal Makhani", "Aloo Paratha"],
+        "main_nonveg": ["Butter Chicken", "Rogan Josh", "Mutton Biryani"],
+        "desserts": ["Gulab Jamun", "Kheer"],
+    },
+
+    "telugu": {
+        "region": "Andhra / Telangana",
+        "starters_veg": ["Mirchi Bajji", "Pesarattu"],
+        "starters_nonveg": ["Gongura Chicken Fry", "Chepala Vepudu"],
+        "main_veg": ["Pappu (dal) with rice", "Pesara Pappu", "Gutti Vankaya"],
+        "main_nonveg": ["Andhra Chicken Biryani", "Kodi Pulusu (Chicken Curry)"],
+        "desserts": ["Bobbatlu (Puran Poli)", "Pootharekulu"],
+    },
+
+    "tamil": {
+        "region": "Tamil Nadu",
+        "starters_veg": ["Medu Vada", "Sundal"],
+        "starters_nonveg": ["Chicken 65 (TN style)", "Fish Fry"],
+        "main_veg": ["Idli & Sambar", "Masala Dosa", "Rasam Rice"],
+        "main_nonveg": ["Chettinad Chicken", "Fish Curry (Meen Kuzhambu)"],
+        "desserts": ["Payasam", "Kesari"],
+    },
+
+    "malayalam": {
+        "region": "Kerala",
+        "starters_veg": ["Banana Chips", "Vegetable Uzhunnu Vada"],
+        "starters_nonveg": ["Kerala Fish Fry", "Chicken 65 (Kerala)"],
+        "main_veg": ["Appam with Vegetable Stew", "Puttu and Kadala Curry"],
+        "main_nonveg": ["Kerala Fish Curry", "Beef Ularthiyathu", "Chicken Stew"],
+        "desserts": ["Palada Payasam", "Ada Pradhaman"],
+    },
+
+    "kannada": {
+        "region": "Karnataka",
+        "starters_veg": ["Maddur Vada", "Bonda"],
+        "starters_nonveg": ["Kheema Cutlet"],
+        "main_veg": ["Bisi Bele Bath", "Ragi Mudde", "Akki Roti"],
+        "main_nonveg": ["Mangalorean Chicken Curry", "Neer Dosa with Fish Curry"],
+        "desserts": ["Mysore Pak", "Kesari"],
+    },
+
+    "bengali": {
+        "region": "West Bengal",
+        "starters_veg": ["Vegetable Chop"],
+        "starters_nonveg": ["Fish Kabiraji", "Kolkata-style Chicken Pakora"],
+        "main_veg": ["Shukto", "Cholar Dal with Luchi"],
+        "main_nonveg": ["Machher Jhol (Fish Curry)", "Ilish Bhapa (Hilsa)"],
+        "desserts": ["Mishti Doi", "Rasgulla"],
+    },
+
+    # Jharkhand-specific mapping (use label "Jharkhand" -> "jharkhand")
+    "jharkhand": {
+        "region": "Jharkhand",
+        "starters_veg": ["Dhuska", "Chilka Roti"],
+        "starters_nonveg": ["Sohari Chicken Fry", "Kodo Chicken", "Jhal Murg"],
+        "main_veg": ["Thekua with Chana Sabzi", "Litti Chokha"],
+        "main_nonveg": [
+            "Chicken Jhol",
+            "Mutton with Kodo/Kutki Millet",
+            "Bamboo Shoot Chicken",
+        ],
+        "desserts": ["Dudh Pitha", "Gur ki Roti"],
+    },
+
+    "gujarati": {
+        "region": "Gujarat",
+        "starters_veg": ["Dhokla", "Khandvi"],
+        "starters_nonveg": ["(Typically veg cuisine; pick local non-veg if needed)"],
+        "main_veg": ["Undhiyu", "Khichdi", "Thepla"],
+        "main_nonveg": ["(Typically veg cuisine; pick local non-veg if needed)"],
+        "desserts": ["Basundi", "Shrikhand"],
+    },
+
+    # default fallback
+    "default": {
+        "region": "Unknown / Other",
+        "starters_veg": ["Local vegetarian starters"],
+        "starters_nonveg": ["Local non-veg starters"],
+        "main_veg": ["Local vegetarian mains"],
+        "main_nonveg": ["Local non-veg mains"],
+        "desserts": ["Local desserts"],
+    },
+}
+
+# NEW: map HuBERT / MFCC labels (state names etc.) to cuisine keys above
+LABEL_ALIAS = {
+    # Andhra / Telangana
+    "andhra_pradesh": "telugu",
+    "ap": "telugu",
+    "andhra": "telugu",
+    "telugu": "telugu",
+    "telangana": "telugu",
+
+    # Kerala
+    "kerala": "kerala",
+    "malayalam": "malayalam",
+
+    # Tamil Nadu
+    "tamil_nadu": "tamil",
+    "tamil": "tamil",
+
+    # Karnataka
+    "karnataka": "kannada",
+    "kannada": "kannada",
+
+    # Jharkhand
+    "jharkhand": "jharkhand",
+    "jharkhand_state": "jharkhand",
+
+    # Gujarat
+    "gujarat": "gujarati",
+    "gujarati": "gujarati",
+
+    # West Bengal
+    "west_bengal": "bengali",
+    "bengal": "bengali",
+    "bengali": "bengali",
+
+    # North India / Hindi belt
+    "hindi": "hindi",
+    "north_india": "hindi",
+    "delhi": "hindi",
+}
+
+
+def format_cuisine_output(pred_label: str) -> str:
+    """Format cuisine recommendations as a readable multiline string."""
+    key = pred_label.lower().strip()
+    key = key.replace(" ", "_")           # handle "Andhra Pradesh" → "andhra_pradesh"
+    key = LABEL_ALIAS.get(key, key)       # map state-style labels → cuisine keys
+
+    info = CUISINE_MAP.get(key, CUISINE_MAP["default"])
+
+    lines = []
+    lines.append(f"Inferred region: {info['region']}")
+    lines.append("")
+    lines.append("Starters (Veg): " + ", ".join(info["starters_veg"]))
+    lines.append("Starters (Non-Veg): " + ", ".join(info["starters_nonveg"]))
+    lines.append("")
+    lines.append("Main Course (Veg): " + ", ".join(info["main_veg"]))
+    lines.append("Main Course (Non-Veg): " + ", ".join(info["main_nonveg"]))
+    lines.append("")
+    lines.append("Desserts: " + ", ".join(info["desserts"]))
+    return "\n".join(lines)
+
+
+def ui_predict_multi(audio_file, feature_choice, hubert_layer_idx, use_trained_model):
+    """Main function used by Gradio UI."""
+    audio_path = normalize_audio_path(audio_file)
+    if audio_path is None:
+        return "No audio provided", ""
+
+    feat = "hubert" if feature_choice == "HuBERT" else "mfcc"
+
+    # If using trained model, pick the correct model info
+    if use_trained_model:
+        model_info = app_state.get("models", {}).get(feat)
+        if not model_info:
+            return (
+                f"No trained {feat.upper()} model available on server. "
+                "Train it or uncheck 'Use trained model'.",
+                "",
+            )
+
+        clf = model_info["clf"]
+        scaler = model_info["scaler"]
+        le = model_info["le"]
+
+        if feat == "hubert" and hubert_layer_idx != model_info.get(
+            "layer", hubert_layer_idx
+        ):
+            return (
+                f"Note: the HuBERT model was trained on layer "
+                f"{model_info.get('layer')}. Set the slider to that layer or re-train.",
+                "",
+            )
+    else:
+        # Fallback path: only MFCC fallback is implemented
+        if feat != "mfcc":
+            return (
+                "Fallback training is only available for MFCC. "
+                "Choose MFCC or upload HuBERT artifacts.",
+                "",
+            )
+
+        # Try to train a quick MFCC model using local data/ folder
+        wav_paths = []
+        if Path("data").exists():
+            wav_paths = [str(p) for p in Path("data").rglob("*.wav")]
+
+        model_info = train_fast_mfcc_fallback(wav_paths[:200])
+        if model_info is None:
+            return (
+                "No fallback MFCC model could be trained (not enough data). "
+                "Upload model artifacts or provide a data/ folder.",
+                "",
+            )
+
+        clf = model_info["clf"]
+        scaler = model_info["scaler"]
+        le = model_info["le"]
+
+    # Extract features
+    try:
+        if feat == "mfcc":
+            x = extract_mfcc_pooled(audio_path)
+        else:
+            x = get_hubert_layer_embedding(audio_path, layer_idx=hubert_layer_idx)
+    except Exception as e:
+        return f"Feature extraction failed: {e}", ""
+
+    # Predict
+    try:
+        Xs = scaler.transform(x.reshape(1, -1))
+        pred_idx = clf.predict(Xs)[0]
+        pred_label = le.inverse_transform([pred_idx])[0]
+    except Exception as e:
+        return f"Prediction failed (shape mismatch or model error): {e}", ""
+
+    # Format cuisine recommendations
+    cuisine_text = format_cuisine_output(pred_label)
+    return pred_label, cuisine_text
+
+
+# Build Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("## Accent Detection + Cuisine Recommendation Demo")
+    gr.Markdown(f"[📄 Open Project Brief]({PROJECT_BRIEF_LOCAL})")
+
+    with gr.Row():
+        audio_in = gr.Audio(
+            type="filepath",
+            label="Upload audio (.wav/.flac/.mp3)",
+        )
+
+        with gr.Column():
+            feature_choice = gr.Radio(
+                ["MFCC", "HuBERT"],
+                value="MFCC",
+                label="Feature Type",
+            )
+            hubert_layer_idx = gr.Slider(
+                minimum=0,
+                maximum=24,
+                value=11,
+                step=1,
+                label="HuBERT Layer",
+            )
+            use_trained_model = gr.Checkbox(
+                value=True,
+                label="Use trained model from notebook",
+            )
+            btn = gr.Button("Predict & Recommend")
+
+    output_label = gr.Textbox(label="Predicted Accent")
+    output_cuisines = gr.Textbox(
+        label="Recommended Cuisines (starters, mains, desserts separated)",
+        lines=12,
+    )
+
+    btn.click(
+        ui_predict_multi,
+        inputs=[audio_in, feature_choice, hubert_layer_idx, use_trained_model],
+        outputs=[output_label, output_cuisines],
+    )
+
+demo.launch(share=True)
+