Audio API files

FastAPI-Audio/client.py

+import requests
+
+# server url
+URL = "http://127.0.0.1:5050/predict"
+
+
+# audio file send for predicting keyword
+FILE_PATH = "audio/fold1/100263-2-0-117.wav"
+
+
+if __name__ == "__main__":
+
+    # open files
+    file = open(FILE_PATH, "rb")
+
+    # package stuff to send and perform POST request
+    values = {"file": (FILE_PATH, file, "audio/fold1/100263-2-0-117.wav")}
+    response = requests.post(URL, files=values)
+    data = response.json()
+
+    print("Predicted keyword: {}".format(data["keyword"]))
\ No newline at end of file

FastAPI-Audio/modelFin.py

+import librosa
+import tensorflow as tf
+import numpy as np
+
+SAVED_MODEL_PATH = "model_weights_acc_new.h5"
+SAMPLES_TO_CONSIDER = 22050
+
+class _modelFin:
+
+    model = None
+    _mapping = [
+        "children_playing",
+        "children_playing2",
+        "children_playing3",
+        "children_playing4",
+        "children_playing5"
+    ]
+
+    _instance = None
+
+
+    def predict(self, file_path):
+        
+        # extract MFCC
+        MFCCs = self.preprocess(file_path)
+
+        # 4-dim array to feed to the model for prediction: (# samples, # time steps, # coefficients, 1)
+        MFCCs = MFCCs[np.newaxis, ..., np.newaxis]
+
+        # get the predicted label
+        predictions = self.model.predict(MFCCs)
+        predicted_index = np.argmax(predictions)
+        predicted_keyword = self._mapping[predicted_index]
+        return predicted_keyword
+
+
+    def preprocess(self, file_path, num_mfcc=40, n_fft=2048, hop_length=512):
+        
+        # load audio file
+        signal, sample_rate = librosa.load(file_path)
+
+        if len(signal) >= SAMPLES_TO_CONSIDER:
+            # ensure consistency of the length of the signal
+            signal = signal[:SAMPLES_TO_CONSIDER]
+
+            # extract MFCCs
+            MFCCs = librosa.feature.mfcc(signal, sample_rate, n_mfcc=num_mfcc, n_fft=n_fft,
+                                         hop_length=hop_length)
+        return MFCCs.T
+
+def modelFin():
+    
+    # ensure an instance is created only the first time the factory function is called
+    if _modelFin._instance is None:
+        _modelFin._instance = _modelFin()
+        _modelFin.model = tf.keras.models.load_model(SAVED_MODEL_PATH)
+    return _modelFin._instance
+
+
+if __name__ == "__main__":
+
+    # create 2 instances of service
+    kss = modelFin()
+    kss1 = modelFin()
+
+    # check that different instances of the service point back to the same object
+    assert kss is kss1
+
+    # make a prediction
+    keyword = kss.predict("100263-2-0-117.wav")
+    print(keyword)
\ No newline at end of file

FastAPI-Audio/server.py

+from fastapi import FastAPI
+from fastapi import UploadFile, File
+import uvicorn
+from fastapi.responses import StreamingResponse
+import random
+import os
+from modelFin import modelFin 
+
+from pydantic import BaseModel
+
+app = FastAPI()
+
+# @app.on_event("startup")
+# def load_model():
+#     global model
+#     model = pickle.load(open("model_tree.pkl", "rb"))
+
+@app.get('/index')
+def hello_world(name:str):
+    return f"Hello {name}!"
+
+@app.post('/audio/predict')
+def predict_audio(file:UploadFile = File(...)):
+
+	#file_name = str(random.randint(0, 100000))
+	#file.save(file_name)
+
+    mod = modelFin()
+    pre_key = modelFin.predict(file)
+
+    received = file.dict()
+    normal = received['children_playing']
+    label2 = received['children_playing2']
+    label3 = received['children_playing3']
+    label4 = received['children_playing4']
+    label5 = received['children_playing5']
+    
+    pred_name = modelFin.predict([[normal, label2, label3,
+                                label4, label5]]).tolist()[0]
+    return {'prediction': pred_name}
+
+
+if __name__ == "__main__":
+    uvicorn.run(app, port=8080, host='0.0.0.0')   
+
+