Add files via upload

Author: jibrahim80

PythonCode/InferSpeechCommands.py

@@ -0,0 +1,46 @@
+from speechCommandModel import *
+import numpy as np
+import librosa
+
+# Load pretrained model
+model = CNN()
+model.load_state_dict(torch.load('model_weights.pth'))
+model.eval()
+
+# Read audio
+sr = 16e3
+samples, sr = librosa.load(filename, sr)
+
+# Define Mel spectrogram arguments
+args = {
+  "n_fft": 512,
+  "hop_length": 160,
+  "win_length":512,
+  "window": "hann",
+  "center":False,
+  "n_mels": 50,
+  "norm": "slaney",
+  "sr":sr,
+  "power":2,
+  "htk": True,
+  "fmin": 0,
+  "fmax": sr/2
+}
+
+# Extract mel spectrogram
+z = librosa.feature.melspectrogram(samples, sr=sr, n_fft=args["n_fft"], hop_length=args["hop_length"], 
+                                   win_length=args["win_length"],window=args["window"], 
+                                   n_mels=args["n_mels"], center=args["center"],
+                                   norm=args["norm"], htk=args["htk"], power=args["power"])
+mel_spectrogram = np.log10(z + 1e-6)
+mel_spectrogram = np.transpose(mel_spectrogram)
+mel_spectrogram2 = mel_spectrogram[np.newaxis,np.newaxis,:, :]
+
+# Pass spectrogram to network
+z = model(torch.FloatTensor(mel_spectrogram2))
+_, yhat = torch.max(z.data, 1)
+
+CLASSES = 'unknown, yes, no, up, down, left, right, on, off, stop, go'.split(', ')
+print("Recognized command: " + CLASSES[yhat])
+
+z = z.cpu().detach().numpy()

PythonCode/convertModelToONNX.py

@@ -0,0 +1,17 @@
+from speechCommandModel import *
+
+# Load pretrained model
+model = CNN()
+model.load_state_dict(torch.load('model_weights.pth'))
+model.eval()
+
+# Convert to ONNX format
+x = torch.empty(1, 1, 98, 50)
+torch.onnx.export(model,
+                      x,
+                      "cmdRecognitionPyTorch.onnx",
+                      export_params=True,
+                      opset_version=9,
+                      do_constant_folding=True,
+                      input_names=['input'],
+                      output_names=['output'])

PythonCode/model_weights.pth

@@ -0,0 +1,81 @@
+import torch.nn as nn
+import torch
+
+NumF = 12
+numHops = 98
+timePoolSize = 13
+dropoutProb = 0.2
+numClasses = 11
+
+class CNN(nn.Module):
+
+    # Contructor
+    def __init__(self, out_1=NumF):
+        super(CNN, self).__init__()
+        self.cnn1 = nn.Conv2d(in_channels=1, out_channels=out_1, kernel_size=3, padding=1)
+        self.batch1 = nn.BatchNorm2d(out_1)
+        self.relu1 = nn.ReLU()
+
+        self.maxpool1 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+
+        self.cnn2 = nn.Conv2d(in_channels=out_1, out_channels=2 * out_1, kernel_size=3, padding=1)
+        self.batch2 = nn.BatchNorm2d(2 * out_1)
+        self.relu2 = nn.ReLU()
+
+        self.maxpool2 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+
+        self.cnn3 = nn.Conv2d(in_channels=2 * out_1, out_channels=4 * out_1, kernel_size=3, padding=1)
+        self.batch3 = nn.BatchNorm2d(4 * out_1)
+        self.relu3 = nn.ReLU()
+
+        self.maxpool3 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+
+        self.cnn4 = nn.Conv2d(in_channels=4 * out_1, out_channels=4 * out_1, kernel_size=3, padding=1)
+        self.batch4 = nn.BatchNorm2d(4 * out_1)
+        self.relu4 = nn.ReLU()
+        self.cnn5 = nn.Conv2d(in_channels=4 * out_1, out_channels=4 * out_1, kernel_size=3, padding=1)
+        self.batch5 = nn.BatchNorm2d(4 * out_1)
+        self.relu5 = nn.ReLU()
+
+        self.maxpool4 = nn.MaxPool2d(kernel_size=(timePoolSize, 1))
+
+        self.dropout = nn.Dropout2d(dropoutProb)
+
+        self.fc = nn.Linear(336, numClasses)
+
+    # Prediction
+    def forward(self, x):
+
+        out = self.cnn1(x)
+        out = self.batch1(out)
+        out = self.relu1(out)
+
+        out = self.maxpool1(out)
+
+        out = self.cnn2(out)
+        out = self.batch2(out)
+        out = self.relu2(out)
+
+        out = self.maxpool2(out)
+
+        out = self.cnn3(out)
+        out = self.batch3(out)
+        out = self.relu3(out)
+
+        out = self.maxpool3(out)
+
+        out = self.cnn4(out)
+        out = self.batch4(out)
+        out = self.relu4(out)
+        out = self.cnn5(out)
+        out = self.batch5(out)
+        out = self.relu5(out)
+
+        out = self.maxpool4(out)
+
+        out = self.dropout(out)
+
+        out = out.view(out.size(0), -1)
+        out = self.fc(out)
+
+        return out