Merge branch 'it18257632' into 'master'

Create keyword spotting service

See merge request !63

backend/IT18257632/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 = [
+        "bed",
+        "bird",
+        "dog",
+        "eight",
+        "go",
+        "left",
+        "marvin",
+        "nine",
+        "no",
+        "one",
+        "seven",
+        "sheila",
+        "tree",
+        "wow",
+        "yes"
+    ]
+    _instance = None
+
+    def predict(self, 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
+
+        new_file_path = file_path.split("/")
+
+        new_file_paths = new_file_path[1].split(".")
+
+        print(f"1-{predicted_keyword}")
+        print(f"2-{new_file_paths[0]}")
+
+        if predicted_keyword == new_file_paths[0]:
+            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/1.wav")
+    # keyword2 = kss.predict("test/blu7.wav")
+
+    print(f"Predicted Keywords: {keyword1}")
+    # print(f"Predicted Keywords: {keyword2}")
\ No newline at end of file