changes

API/model/colorModel.py → API/model/color/colorModel.py

-from flask import jsonify
+import os
+import random
+from flask import jsonify, Flask, request
 from API.db.dbConnection import get_all_data, insert_data_json, insert_data, insert, get_data
+from API.model.color.keyword_spotting_service import Keyword_Spotting_service
 from API.util.util import getUUID
 
 # get color activity 1 fun
@@ -86,4 +89,43 @@ def save_color_session_details(userId, status):
     qry = 'INSERT INTO colorsession (id,userId,token,status) VALUES (NULL, %s, %s, %s)'
     args = (userId, token, status)
     insert(qry, args)
-    return token
\ No newline at end of file
+    return token
+
+#predict audio file
+def sendPredictData(audioFile, name):
+
+    # print(f"{name}")
+
+    path = "color/test/green.wav"
+
+    # print(f"{path}")
+
+    audio_file = open(path, "rb")
+
+    values = {
+        "file": (path, audio_file, "audio/wav")
+    }
+    audio_file = values["file"]
+    file_name = str(random.randint(0, 100000))
+    audio_file.save((file_name))
+    # print(f"{path}")
+
+    # get file name
+    predict_file_name = audio_file.filename
+    new_predict_file_name = predict_file_name.split(".")
+    FPFN = new_predict_file_name[0]
+
+    # invoke keyword spotting service
+    kss = Keyword_Spotting_service()
+
+    # make a prediction
+    predicted_keyword = kss.predict(file_name, FPFN)
+
+    # remove the audio file
+    os.remove(file_name)
+
+    # send back the predicted keyword in json format
+    data = {"Keyword": predicted_keyword}
+
+    print(f"data: {data}")
+    return jsonify(data)
\ No newline at end of file

API/model/color/keyword_spotting_service.py

+import tensorflow.keras as keras
+import numpy as np
+import librosa
+
+MODEL_PATH = "model.h5"
+NUM_SAMPLES_TO_CONSIDER = 22050 # 1 sec
+
+class _Keyword_Spotting_Service:
+
+    model = None
+    _mappings = [
+        "black",
+        "blue",
+        "green",
+        "red",
+        "white",
+        "yellow"
+    ]
+    _instance = None
+    # 78293
+    def predict(self, file_path, FPFN):
+
+        print(f"{file_path}")
+
+        # extract MFCCs
+        MFCCs = self.preprocess(file_path) # (# segment, # coefficients)
+
+        # convert 2d MFCCs array into 4d array -> (# samples, # segment, # coefficients, # channels)
+        MFCCs = MFCCs[np.newaxis, ..., np.newaxis]
+
+        # make prediction
+        predictions = self.model.predict(MFCCs)
+        predicted_index = np.argmax(predictions)
+        predicted_keyword = self._mappings[predicted_index]
+
+        # return predicted_keyword
+
+        if predicted_keyword == FPFN:
+            return predicted_keyword
+        else:
+            return "No Prediction"
+
+    def preprocess(self, file_path, n_mfcc=13, n_fft=2048, hop_length=512):
+
+        # load audio file
+        signal, sr = librosa.load(file_path)
+
+        # ensure consistency in the audio file length
+        if len(signal) > NUM_SAMPLES_TO_CONSIDER:
+            signal = signal[:NUM_SAMPLES_TO_CONSIDER]
+
+        # extract MFCCs
+        MFCCs = librosa.feature.mfcc(signal, n_mfcc=n_mfcc, n_fft=n_fft, hop_length=hop_length)
+
+        return MFCCs.T
+
+def Keyword_Spotting_service():
+
+    # ensure that we only have 1 instance of KSS
+    if _Keyword_Spotting_Service._instance is None:
+        _Keyword_Spotting_Service._instance = _Keyword_Spotting_Service()
+        _Keyword_Spotting_Service.model = keras.models.load_model(MODEL_PATH)
+
+    return _Keyword_Spotting_Service._instance
+
+if __name__ == "__main__":
+    kss = Keyword_Spotting_service()
+
+    keyword1 = kss.predict("test/black.wav")
+    keyword2 = kss.predict("test/blue.wav")
+    keyword3 = kss.predict("test/green.wav")
+    keyword4 = kss.predict("test/red.wav")
+    keyword5 = kss.predict("test/white.wav")
+    keyword6 = kss.predict("test/yellow.wav")
+    keyword7 = kss.predict("test/other.wav")
+
+    print(f"Predicted Keywords: {keyword1}")
+    print(f"Predicted Keywords: {keyword2}")
+    print(f"Predicted Keywords: {keyword3}")
+    print(f"Predicted Keywords: {keyword4}")
+    print(f"Predicted Keywords: {keyword5}")
+    print(f"Predicted Keywords: {keyword6}")
+    print(f"Predicted Keywords: {keyword7}")
\ No newline at end of file