initial commit
Browse files- MODEL_CARD +15 -0
- kronk.json +16 -0
- kronk.tflite +3 -0
- model_summary.txt +103 -0
- tflite_streaming_roc.txt +11 -0
MODEL_CARD
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Model card for 'Hey, Kronk!' microwakeword
|
| 2 |
+
|
| 3 |
+
* Language: en
|
| 4 |
+
|
| 5 |
+
## Dataset
|
| 6 |
+
|
| 7 |
+
Dataset generated by Piper-TTS using the basic training notebook from:
|
| 8 |
+
|
| 9 |
+
https://github.com/kahrendt/microWakeWord/blob/main/notebooks/basic_training_notebook.ipynb
|
| 10 |
+
|
| 11 |
+
Noise and negative training data also from above.
|
| 12 |
+
|
| 13 |
+
## Training
|
| 14 |
+
|
| 15 |
+
Trained using default notebook settings.
|
kronk.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"type": "micro",
|
| 3 |
+
"wake_word": "Hey Kronk",
|
| 4 |
+
"author": "russdill",
|
| 5 |
+
"website": "https://huggingface.co/russdill/microwakeword-kronk",
|
| 6 |
+
"model": "https://huggingface.co/russdill/microwakeword-kronk/resolve/main/kronk.tflite",
|
| 7 |
+
"version": 2,
|
| 8 |
+
"trained_languages": ["en"],
|
| 9 |
+
"micro": {
|
| 10 |
+
"probability_cutoff": 0.99,
|
| 11 |
+
"sliding_window_size": 5,
|
| 12 |
+
"feature_step_size": 10,
|
| 13 |
+
"tensor_arena_size": 25552,
|
| 14 |
+
"minimum_esphome_version": "2024.7.0"
|
| 15 |
+
}
|
| 16 |
+
}
|
kronk.tflite
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:335c512ac6bc80c0f9c6d9c91a20d1a81896f98e82af31c601d0b81b2d2c9642
|
| 3 |
+
size 60944
|
model_summary.txt
ADDED
|
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model: "functional"
|
| 2 |
+
βββββββββββββββββββββββββββββββββ³ββββββββββββββββββββββββββββ³ββββββββββββββββββ³βββββββββββββββββββββββββββββ
|
| 3 |
+
β Layer (type) β Output Shape β Param # β Connected to β
|
| 4 |
+
β‘βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
|
| 5 |
+
β input_layer (InputLayer) β (128, 204, 40) β 0 β - β
|
| 6 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 7 |
+
β expand_dims (ExpandDims) β (128, 204, 1, 40) β 0 β input_layer[0][0] β
|
| 8 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 9 |
+
β stream (Stream) β (128, 67, 1, 32) β 6,400 β expand_dims[0][0] β
|
| 10 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 11 |
+
β activation (Activation) β (128, 67, 1, 32) β 0 β stream[0][0] β
|
| 12 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 13 |
+
β stream_1 (Stream) β (128, 67, 1, 32) β 0 β activation[0][0] β
|
| 14 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 15 |
+
β depthwise_conv2d β (128, 63, 1, 32) β 192 β stream_1[0][0] β
|
| 16 |
+
β (DepthwiseConv2D) β β β β
|
| 17 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 18 |
+
β conv2d_1 (Conv2D) β (128, 63, 1, 64) β 2,048 β depthwise_conv2d[0][0] β
|
| 19 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 20 |
+
β batch_normalization β (128, 63, 1, 64) β 256 β conv2d_1[0][0] β
|
| 21 |
+
β (BatchNormalization) β β β β
|
| 22 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 23 |
+
β activation_1 (Activation) β (128, 63, 1, 64) β 0 β batch_normalization[0][0] β
|
| 24 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 25 |
+
β stream_2 (Stream) β (128, 63, 1, 64) β 0 β activation_1[0][0] β
|
| 26 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 27 |
+
β channel_split (ChannelSplit) β [(128, 63, 1, 32), (128, β 0 β stream_2[0][0] β
|
| 28 |
+
β β 63, 1, 32)] β β β
|
| 29 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 30 |
+
β strided_keep (StridedKeep) β (128, 63, 1, 32) β 0 β channel_split[0][0] β
|
| 31 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 32 |
+
β strided_keep_1 (StridedKeep) β (128, 63, 1, 32) β 0 β channel_split[0][1] β
|
| 33 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 34 |
+
β depthwise_conv2d_1 β (128, 57, 1, 32) β 256 β strided_keep[0][0] β
|
| 35 |
+
β (DepthwiseConv2D) β β β β
|
| 36 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 37 |
+
β depthwise_conv2d_2 β (128, 53, 1, 32) β 384 β strided_keep_1[0][0] β
|
| 38 |
+
β (DepthwiseConv2D) β β β β
|
| 39 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 40 |
+
β strided_drop (StridedDrop) β (128, 53, 1, 32) β 0 β depthwise_conv2d_1[0][0] β
|
| 41 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 42 |
+
β strided_drop_1 (StridedDrop) β (128, 53, 1, 32) β 0 β depthwise_conv2d_2[0][0] β
|
| 43 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 44 |
+
β concatenate (Concatenate) β (128, 53, 1, 64) β 0 β strided_drop[0][0], β
|
| 45 |
+
β β β β strided_drop_1[0][0] β
|
| 46 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 47 |
+
β conv2d_2 (Conv2D) β (128, 53, 1, 64) β 4,096 β concatenate[0][0] β
|
| 48 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 49 |
+
β batch_normalization_1 β (128, 53, 1, 64) β 256 β conv2d_2[0][0] β
|
| 50 |
+
β (BatchNormalization) β β β β
|
| 51 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 52 |
+
β activation_2 (Activation) β (128, 53, 1, 64) β 0 β batch_normalization_1[0][β¦ β
|
| 53 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 54 |
+
β stream_3 (Stream) β (128, 53, 1, 64) β 0 β activation_2[0][0] β
|
| 55 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 56 |
+
β channel_split_1 β [(128, 53, 1, 32), (128, β 0 β stream_3[0][0] β
|
| 57 |
+
β (ChannelSplit) β 53, 1, 32)] β β β
|
| 58 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 59 |
+
β strided_keep_2 (StridedKeep) β (128, 53, 1, 32) β 0 β channel_split_1[0][0] β
|
| 60 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 61 |
+
β strided_keep_3 (StridedKeep) β (128, 53, 1, 32) β 0 β channel_split_1[0][1] β
|
| 62 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 63 |
+
β depthwise_conv2d_3 β (128, 45, 1, 32) β 320 β strided_keep_2[0][0] β
|
| 64 |
+
β (DepthwiseConv2D) β β β β
|
| 65 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 66 |
+
β depthwise_conv2d_4 β (128, 39, 1, 32) β 512 β strided_keep_3[0][0] β
|
| 67 |
+
β (DepthwiseConv2D) β β β β
|
| 68 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 69 |
+
β strided_drop_2 (StridedDrop) β (128, 39, 1, 32) β 0 β depthwise_conv2d_3[0][0] β
|
| 70 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 71 |
+
β strided_drop_3 (StridedDrop) β (128, 39, 1, 32) β 0 β depthwise_conv2d_4[0][0] β
|
| 72 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 73 |
+
β concatenate_1 (Concatenate) β (128, 39, 1, 64) β 0 β strided_drop_2[0][0], β
|
| 74 |
+
β β β β strided_drop_3[0][0] β
|
| 75 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββοΏ½οΏ½οΏ½βββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 76 |
+
β conv2d_3 (Conv2D) β (128, 39, 1, 64) β 4,096 β concatenate_1[0][0] β
|
| 77 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 78 |
+
β batch_normalization_2 β (128, 39, 1, 64) β 256 β conv2d_3[0][0] β
|
| 79 |
+
β (BatchNormalization) β β β β
|
| 80 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 81 |
+
β activation_3 (Activation) β (128, 39, 1, 64) β 0 β batch_normalization_2[0][β¦ β
|
| 82 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 83 |
+
β stream_4 (Stream) β (128, 39, 1, 64) β 0 β activation_3[0][0] β
|
| 84 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 85 |
+
β depthwise_conv2d_5 β (128, 17, 1, 64) β 1,536 β stream_4[0][0] β
|
| 86 |
+
β (DepthwiseConv2D) β β β β
|
| 87 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 88 |
+
β conv2d_4 (Conv2D) β (128, 17, 1, 64) β 4,096 β depthwise_conv2d_5[0][0] β
|
| 89 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 90 |
+
β batch_normalization_3 β (128, 17, 1, 64) β 256 β conv2d_4[0][0] β
|
| 91 |
+
β (BatchNormalization) β β β β
|
| 92 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 93 |
+
β activation_4 (Activation) β (128, 17, 1, 64) β 0 β batch_normalization_3[0][β¦ β
|
| 94 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 95 |
+
β stream_5 (Stream) β (128, 17, 1, 64) β 0 β activation_4[0][0] β
|
| 96 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 97 |
+
β flatten (Flatten) β (128, 1088) β 0 β stream_5[0][0] β
|
| 98 |
+
βββββββββββββββββββββββββββββββββΌββββββββββββββββββββββββββββΌββββββββββββββββββΌβββββββββββββββββββββββββββββ€
|
| 99 |
+
β dense (Dense) β (128, 1) β 1,089 β flatten[0][0] β
|
| 100 |
+
βββββββοΏ½οΏ½οΏ½βββββββββββββββββββββββββ΄ββββββββββββββββββββββββββββ΄ββββββββββββββββββ΄βββββββββββββββββββββββββββββ
|
| 101 |
+
Total params: 26,049 (101.75 KB)
|
| 102 |
+
Trainable params: 25,537 (99.75 KB)
|
| 103 |
+
Non-trainable params: 512 (2.00 KB)
|
tflite_streaming_roc.txt
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
AUC 0.09219
|
| 2 |
+
Cutoff 0.99: frr=0.0500; faph=0.000
|
| 3 |
+
Cutoff 0.98: frr=0.0500; faph=0.562
|
| 4 |
+
Cutoff 0.96: frr=0.0500; faph=0.750
|
| 5 |
+
Cutoff 0.93: frr=0.0500; faph=0.937
|
| 6 |
+
Cutoff 0.91: frr=0.0500; faph=1.125
|
| 7 |
+
Cutoff 0.89: frr=0.0400; faph=1.312
|
| 8 |
+
Cutoff 0.84: frr=0.0400; faph=1.500
|
| 9 |
+
Cutoff 0.80: frr=0.0400; faph=1.687
|
| 10 |
+
Cutoff 0.77: frr=0.0400; faph=1.875
|
| 11 |
+
Cutoff 0.77: frr=0.0400; faph=2.000
|