delete

API/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