Update model

Signed-off-by: Binh Le <[email protected]>

classifier.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1c5cf560dc904388ca459804ce8ee11944b3edc102a146d2a7c3c3ea1b3b67b
-size 9067
+oid sha256:420ed305f44ddf60cd109627682c75c98b76a45c15b0cd43d44606a867137289
+size 5995

embedding_model.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0766e6aba5bd68e4a080116bb8ae17afc616284d1c2a50a22d67202b56e73b11
-size 1065702
+oid sha256:85c18d1b214d7bb56ab3748e45ba339e0893f9baeb516ebb7e65aeec7b540123
+size 798438

hyperparams.yaml

@@ -9,7 +9,7 @@ pretrained_path: /content/drive/MyDrive/audify/inference
 n_mels: 80
 
 # Output parameters
-out_n_neurons: 4 # Possible languages in the dataset
+out_n_neurons: 5 # Possible languages in the dataset
 
 # Model params
 compute_features: !new:speechbrain.lobes.features.Fbank
@@ -24,7 +24,7 @@ CNN: !new:speechbrain.lobes.models.convolution.ConvolutionFrontEnd
   input_shape: (null, null, 80)
   num_blocks: 3
   num_layers_per_block: 1
-  out_channels: (128, 256, 512)
+  out_channels: (128, 256, 256)
   kernel_sizes: (3, 3, 1)
   strides: (2, 2, 1)
   residuals: (False, False, False)
@@ -42,7 +42,7 @@ embedding_model: !new:speechbrain.nnet.containers.LengthsCapableSequential
   pooling: !ref <pooling>
 
 classifier: !new:speechbrain.lobes.models.ECAPA_TDNN.Classifier
-  input_size: 512
+  input_size: 256
   out_neurons: !ref <out_n_neurons>
 
 modules:

label_encoder.txt

@@ -1,6 +1,7 @@
-'speech' => 0
-'noisy_speech' => 1
-'corrupted' => 2
+'clean' => 0
+'noisy' => 1
+'corrupt' => 2
 'music' => 3
+'background' => 4
 ================
 'starting_index' => 0