Upload 6 files

app.py

@@ -0,0 +1,74 @@
+# ======================
+# app.py
+# ======================
+
+from flask import Flask, request, jsonify
+import tensorflow as tf
+import numpy as np
+import pandas as pd
+import librosa
+from transformers import AutoTokenizer, TFAutoModel
+
+# Load saved files
+model = tf.keras.models.load_model("model.h5")
+scaler = pd.read_pickle("scaler.pkl")
+encoder = pd.read_pickle("label_encoder.pkl")
+meta = pd.read_excel("raga_metadata.xlsx")
+
+# Load IndicBERT model
+tokenizer = AutoTokenizer.from_pretrained("ai4bharat/IndicBERTv2-MLM-only")
+bert_model = TFAutoModel.from_pretrained("ai4bharat/IndicBERTv2-MLM-only", from_pt=True)
+
+app = Flask(__name__)
+
+def extract_features(file_path):
+    y, sr = librosa.load(file_path, sr=22050)
+    features = {
+        "chroma_stft": np.mean(librosa.feature.chroma_stft(y=y, sr=sr)),
+        "spec_cent": np.mean(librosa.feature.spectral_centroid(y=y, sr=sr)),
+    }
+    mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=18)
+    for i in range(18):
+        features[f"mfcc{i+1}"] = np.mean(mfccs[i])
+    return pd.DataFrame([features])
+
+def predict_raga(audio_df, raga_name):
+    audio_scaled = scaler.transform(audio_df)
+    audio_lstm_input = audio_scaled.reshape((1, 1, audio_scaled.shape[1]))
+
+    # Get description for this raga
+    description_text = meta[meta['raga'] == raga_name]['description'].values
+    if len(description_text) == 0:
+        description_text = [""]
+
+    desc_tok = tokenizer(description_text.tolist(), padding=True, truncation=True, max_length=64, return_tensors="tf")
+    desc_embed = bert_model(desc_tok['input_ids'], attention_mask=desc_tok['attention_mask'])[0][:, 0, :]
+
+    pred = model.predict([audio_lstm_input, desc_embed])
+    return encoder.inverse_transform([np.argmax(pred)])[0]
+
+@app.route("/")
+def home():
+    return "🎶 Raga Prediction API is Live!"
+
+@app.route("/predict", methods=["POST"])
+def predict():
+    try:
+        audio_file = request.files['audio']
+        raga_name = request.form['raga_name']
+
+        temp_audio_path = "temp_audio.wav"
+        audio_file.save(temp_audio_path)
+
+        features = extract_features(temp_audio_path)
+        predicted_raga = predict_raga(features, raga_name)
+
+        return jsonify({
+            "predicted_raga": predicted_raga
+        })
+
+    except Exception as e:
+        return jsonify({"error": str(e)})
+
+if __name__ == "__main__":
+    app.run(host="0.0.0.0", port=7860)

label_encoder.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9527a080b81de4f169669257ca18c6a7240a235f5e18c0a1c5d09414b89f1afa
+size 1035

raga_predictor_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12cfb239bef9460e0c13f0a1466b36f058fa3685d640d4b8a0aa2b863fa20ad5
+size 3708664

requirements.txt

@@ -0,0 +1,11 @@
+tensorflow
+keras
+transformers
+scikit-learn
+librosa
+lime
+pandas
+numpy
+matplotlib
+openpyxl
+flask

scaler.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cb12ac67332836dd372b4ddff48648ac39e9e46217da55e32ee10c63d0ad2e1
+size 1655