KWS models - update

README.md

@@ -1,4 +1,3 @@
-
 # Model Zoo
 ![version](https://img.shields.io/badge/version-21.08-0091BD)
 > A collection of machine learning models optimized for Arm IP.
@@ -109,7 +108,7 @@
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.929</td>
+        <td align="center">0.931</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/cnn_medium/tflite_int8">CNN Medium INT8 *</a></td>
@@ -119,7 +118,7 @@
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.913</td>
+        <td align="center">0.911</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/cnn_small/tflite_int8">CNN Small INT8 *</a></td>
@@ -129,7 +128,7 @@
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.914</td>
+        <td align="center">0.912</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/dnn_large/tflite_int8">DNN Large INT8 *</a></td>
@@ -149,7 +148,7 @@
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.846</td>
+        <td align="center">0.844</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/dnn_small/tflite_int8">DNN Small INT8 *</a></td>
@@ -159,7 +158,7 @@
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.827</td>
+        <td align="center">0.825</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/ds_cnn_large/tflite_clustered_fp32">DS-CNN Clustered FP32 *</a></td>
@@ -199,7 +198,7 @@
         <td align="center">:heavy_check_mark: HERO</td>
         <td align="center">:heavy_check_mark: </td>
         <td align="center">:heavy_check_mark: </td>
-        <td align="center">0.934</td>
+        <td align="center">0.941</td>
     </tr>
     <tr>
         <td><a href="models/keyword_spotting/ds_cnn_small/tflite_int8">DS-CNN Small INT8 *</a></td>

models/keyword_spotting/cnn_large/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | 59e6986c3eca496fa3d54176ac66bb7dc9ff36e0 |
-|  Size (Bytes)       | 486560 |
+|  SHA-1 Hash         | a61ab748ae8f52f78ab568342db67a792c6ecf34 |
+|  Size (Bytes)       | 484600 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 
@@ -27,7 +27,7 @@ Dataset: Google Speech Commands Test Set
 
 | Metric | Value |
 |--------|-------|
-| Accuracy | 0.929 |
+| Accuracy | 0.931 |
 
 ## Performance
 | Platform | Optimized |

models/keyword_spotting/cnn_large/tflite_int8/definition.yaml

@@ -1,19 +1,20 @@
 benchmark:
   Google Speech Commands test set:
-    Accuracy: 0.929
+    Accuracy: 93.09%
 description: 'This is a fully quantized version (asymmetrical int8) of the CNN Large
   model developed by Arm, with training checkpoints, from the Hello Edge paper. Code
   to recreate this model can be found here: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m'
 license:
 - Apache-2.0
 network:
-  file_size_bytes: 486560
+  file_size_bytes: 484600
   filename: cnn_l_quantized.tflite
   framework: TensorFlow Lite
   hash:
     algorithm: sha1
-    value: 59e6986c3eca496fa3d54176ac66bb7dc9ff36e0
+    value: a61ab748ae8f52f78ab568342db67a792c6ecf34
   provenance: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m
+  quality_level: null
 network_parameters:
   input_nodes:
   - description: The input is a processed MFCCs of shape (1, 490)

models/keyword_spotting/cnn_medium/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | 389c6c2c7d289c0018e2dabcc66271811e52874c |
-|  Size (Bytes)       | 187840 |
+|  SHA-1 Hash         | 6bc68074d960bbb0c695e19fd96fd7903131ef60 |
+|  Size (Bytes)       | 186064 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 
@@ -27,7 +27,7 @@ Dataset: Google Speech Commands Test Set
 
 | Metric | Value |
 |--------|-------|
-| Accuracy | 0.913 |
+| Accuracy | 0.911 |
 
 ## Performance
 | Platform | Optimized |

models/keyword_spotting/cnn_medium/tflite_int8/definition.yaml

@@ -1,19 +1,20 @@
 benchmark:
   Google Speech Commands test set:
-    Accuracy: 0.913
+    Accuracy: 91.08%
 description: 'This is a fully quantized version (asymmetrical int8) of the CNN Medium
   model developed by Arm, with training checkpoints, from the Hello Edge paper. Code
   to recreate this model can be found here: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m'
 license:
 - Apache-2.0
 network:
-  file_size_bytes: 187840
+  file_size_bytes: 186064
   filename: cnn_m_quantized.tflite
   framework: TensorFlow Lite
   hash:
     algorithm: sha1
-    value: 389c6c2c7d289c0018e2dabcc66271811e52874c
+    value: 6bc68074d960bbb0c695e19fd96fd7903131ef60
   provenance: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m
+  quality_level: null
 network_parameters:
   input_nodes:
   - description: The input is a processed MFCCs of shape (1, 490)

models/keyword_spotting/cnn_small/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | d3c8f4b468545d7012383f2a312bef6245a3b599 |
-|  Size (Bytes)       | 76752 |
+|  SHA-1 Hash         | 3415f88dfb8f78fe47d282d68ccbc3ce71a7510f |
+|  Size (Bytes)       | 75400 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 
@@ -27,7 +27,7 @@ Dataset: Google Speech Commands Test Set
 
 | Metric | Value |
 |--------|-------|
-| Accuracy | 0.914 |
+| Accuracy | 0.912 |
 
 ## Performance
 | Platform | Optimized |

models/keyword_spotting/cnn_small/tflite_int8/definition.yaml

@@ -1,19 +1,20 @@
 benchmark:
   Google Speech Commands test set:
-    Accuracy: 0.914
+    Accuracy: 91.23%
 description: 'This is a fully quantized version (asymmetrical int8) of the CNN Small
   model developed by Arm, with training checkpoints, from the Hello Edge paper. Code
   to recreate this model can be found here: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m'
 license:
 - Apache-2.0
 network:
-  file_size_bytes: 76752
+  file_size_bytes: 75400
   filename: cnn_s_quantized.tflite
   framework: TensorFlow Lite
   hash:
     algorithm: sha1
-    value: d3c8f4b468545d7012383f2a312bef6245a3b599
+    value: 3415f88dfb8f78fe47d282d68ccbc3ce71a7510f
   provenance: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m
+  quality_level: null
 network_parameters:
   input_nodes:
   - description: The input is a processed MFCCs of shape (1, 490)

models/keyword_spotting/dnn_large/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | 16e03dda20ae81dfba6a567e6e7563ca67596969 |
-|  Size (Bytes)       | 502928 |
+|  SHA-1 Hash         | 2b1ee34e4c87ba6f24092c7457593227099efaf1 |
+|  Size (Bytes)       | 502272 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 

models/keyword_spotting/dnn_large/tflite_int8/definition.yaml

@@ -1,19 +1,20 @@
 benchmark:
   Google Speech Commands test set:
-    Accuracy: 0.863
+    Accuracy: 86.26%
 description: 'This is a fully quantized version (asymmetrical int8) of the DNN Large
   model developed by Arm, with training checkpoints, from the Hello Edge paper. Code
   to recreate this model can be found here: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m'
 license:
 - Apache-2.0
 network:
-  file_size_bytes: 502928
+  file_size_bytes: 502272
   filename: dnn_l_quantized.tflite
   framework: TensorFlow Lite
   hash:
     algorithm: sha1
-    value: 16e03dda20ae81dfba6a567e6e7563ca67596969
+    value: 2b1ee34e4c87ba6f24092c7457593227099efaf1
   provenance: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m
+  quality_level: null
 network_parameters:
   input_nodes:
   - description: The input is a processed MFCCs of shape (1, 250)

models/keyword_spotting/dnn_medium/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | 57ad3cf78f736819b8897f5de51f7e9a4cbd5689 |
-|  Size (Bytes)       | 204480 |
+|  SHA-1 Hash         | 7e138f99cfc6a603a1fc735a2d9c3e28a41a6a43 |
+|  Size (Bytes)       | 203832 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 
@@ -27,7 +27,7 @@ Dataset: Google Speech Commands Test Set
 
 | Metric | Value |
 |--------|-------|
-| Accuracy | 0.846 |
+| Accuracy | 0.844 |
 
 ## Performance
 | Platform | Optimized |

models/keyword_spotting/dnn_medium/tflite_int8/definition.yaml

@@ -1,19 +1,20 @@
 benchmark:
   Google Speech Commands test set:
-    Accuracy: 0.846
+    Accuracy: 84.44%
 description: 'This is a fully quantized version (asymmetrical int8) of the DNN Medium
   model developed by Arm, with training checkpoints, from the Hello Edge paper. Code
   to recreate this model can be found here: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m'
 license:
 - Apache-2.0
 network:
-  file_size_bytes: 204480
+  file_size_bytes: 203832
   filename: dnn_m_quantized.tflite
   framework: TensorFlow Lite
   hash:
     algorithm: sha1
-    value: 57ad3cf78f736819b8897f5de51f7e9a4cbd5689
+    value: 7e138f99cfc6a603a1fc735a2d9c3e28a41a6a43
   provenance: https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m
+  quality_level: null
 network_parameters:
   input_nodes:
   - description: The input is a processed MFCCs of shape (1, 250)

models/keyword_spotting/dnn_small/tflite_int8/README.md

@@ -17,8 +17,8 @@ Code to recreate this model can be found here: https://github.com/ARM-software/M
 | Network Information |  Value         |
 |---------------------|------------------|
 |  Framework          | TensorFlow Lite |
-|  SHA-1 Hash         | 5b00a7eb54eb2650c50026ddef2b3134a71ab6cf |
-|  Size (Bytes)       | 84192 |
+|  SHA-1 Hash         | 4b92e09fb43b2f042ce2811b91c7c67bf7186b6b |
+|  Size (Bytes)       | 83544 |
 |  Provenance         | https://github.com/ARM-software/ML-examples/tree/master/tflu-kws-cortex-m |
 |  Paper              | https://arxiv.org/abs/1711.07128 |
 
@@ -27,7 +27,7 @@ Dataset: Google Speech Commands Test Set
 
 | Metric | Value |
 |--------|-------|
-| Accuracy | 0.827 |
+| Accuracy | 0.825 |
 
 ## Performance
 | Platform | Optimized |