app.py

# -*- coding: utf-8 -*-
"""SONG.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1D_uRTl2aO65nt7ABxgODlb57aLqY05HC
"""

import gradio as gr
import numpy as np
import librosa
import os
import requests
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, classification_report
import joblib  # to save/load model

# === CONFIGURATION ===
SAAVN_SEARCH_URL = "https://saavn.dev/api/search/songs?query={query}&limit=1"
DATA_DIR = "audio"
  # update this if your folder is different

# Create data directory if it doesn't exist
if not os.path.exists(DATA_DIR):
    os.makedirs(DATA_DIR)
    print(f"Created directory: {DATA_DIR}. Please upload your audio dataset into this folder.")
    # You might want to exit or handle this case where data is missing

# === FEATURE EXTRACTION ===
def extract_features(file_path):
    try:
        audio, sr = librosa.load(file_path, duration=3, offset=0.5)
        mfccs = librosa.feature.mfcc(y=audio, sr=sr, n_mfcc=13)
        return np.mean(mfccs.T, axis=0)
    except Exception as e:
        print("Error loading file:", e)
        return None

# === LOAD DATA AND TRAIN MODEL ===
def train_model(data_dir):
    features = []
    labels = []
    for folder in os.listdir(data_dir):
        emotion = folder.split('_')[-1].lower()
        folder_path = os.path.join(data_dir, folder)
        for filename in os.listdir(folder_path):
            if filename.endswith('.wav'):
                file_path = os.path.join(folder_path, filename)
                mfcc = extract_features(file_path)
                if mfcc is not None:
                    features.append(mfcc)
                    labels.append(emotion)

    # Check if any data was loaded
    if not features:
        print(f"No audio files found in {data_dir}. Please upload your dataset.")
        return None # Or raise an error

    X = np.array(features)
    y = np.array(labels)
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
    model = RandomForestClassifier()
    model.fit(X_train, y_train)

    # Optional: Save model
    joblib.dump(model, "voice_mood_model.pkl")

    # Evaluate (Optional)
    y_pred = model.predict(X_test)
    print("✅ Model Trained - Accuracy:", accuracy_score(y_test, y_pred))
    print(classification_report(y_test, y_pred))
    return model

# Load existing or train new model
model = None # Initialize model to None
if os.path.exists("voice_mood_model.pkl"):
    try:
        model = joblib.load("voice_mood_model.pkl")
        print("Loaded existing model.")
    except Exception as e:
        print(f"Error loading model: {e}. Retraining model.")
        model = train_model(DATA_DIR)
else:
    model = train_model(DATA_DIR)

# Ensure model is trained before proceeding
if model is None:
    print("Model could not be trained or loaded. Exiting.")
    # You might want to add sys.exit() here if running as a script
else:
    # === PREDICT EMOTION FROM AUDIO ===
    def predict_emotion(file_path):
        mfcc = extract_features(file_path)
        if mfcc is not None:
            # Ensure the model is available before predicting
            if model:
                return model.predict(mfcc.reshape(1, -1))[0]
            else:
                print("Model not loaded or trained.")
                return "Error: Model not available"
        return "Unknown"

    # === GET SONG FROM SAAVN ===
    def get_song_from_mood(mood, language="tamil"):
        query = f"{mood} {language} song"
        url = SAAVN_SEARCH_URL.format(query=query)
        try:
            response = requests.get(url)
            response.raise_for_status() # Raise an exception for bad status codes
            data = response.json()
            if data.get("data") and data["data"]["results"]:
                song = data["data"]["results"][0]
                title = song["name"]
                artist = (
                    song.get("artists", {}).get("primary", [{}])[0].get("name") or
                    song.get("primaryArtists") or song.get("artist") or "Unknown Artist"
                )
                return f"🎶 **{title}** by *{artist}*\n🔗 [Listen here]({song['url']})"
        except requests.exceptions.RequestException as e:
            print("Song fetch error:", e)
        except Exception as e:
            print("Song fetch error:", e)
        return "😕 No song found for this mood."

    # === GRADIO INTERFACE FUNCTION ===
    def detect_and_recommend(audio_path):
        if model: # Check if model is available
            mood = predict_emotion(audio_path)
            song = get_song_from_mood(mood)
            return f"🎭 Detected Mood: **{mood}**\n\n{song}"
        else:
            return "Model not trained or loaded. Cannot detect mood."


    # === LAUNCH GRADIO APP ===
    gr.Interface(
        fn=detect_and_recommend,
        inputs=gr.Audio(type="filepath", label="🎙 Upload Your Voice"),
        outputs="markdown",
        title="🎤 Voice2Vibes (Offline Version)",
        description="Detects mood from your voice using your dataset and recommends a matching Tamil song 🎶"
    ).launch()