Edge Impulse Inc. / Tutorial: Responding to your voice Public

EON Tuner

The EON Tuner helps you find the most optimal architecture for your embedded machine-learning application. Clone this project to use the EON Tuner.

Target

No name set

Keyword spotting

Cortex-M7 216MHz

100 ms

340 kB

1024 kB

Filters

Status

DSP type

Network type

View

Data set

Precision

Sort

General

F1-score

Precision

Recall

96%
mfe-conv2d-622
PERFORMANCE
LATENCY
190 ms of 100 ms
Exceeds target by 90 ms
RAM
42 kB of 340 kB
ROM
57 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.032 | 0.016 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 8 3 -
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.5

10/27/2022, 5:28:01 PM

95%
mfcc-conv2d-7a5
PERFORMANCE
LATENCY
192 ms of 100 ms
Exceeds target by 92 ms
RAM
36 kB of 340 kB
ROM
129 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.05 | 0.025 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
conv2d 128 3 -
dropout - - 0.5

10/27/2022, 5:46:10 PM

94%
mfe-conv2d-ef6
PERFORMANCE
LATENCY
217 ms of 100 ms
Exceeds target by 117 ms
RAM
31 kB of 340 kB
ROM
56 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.05 | 0.05 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:35:36 PM

94%
mfcc-conv2d-0e5
PERFORMANCE
LATENCY
196 ms of 100 ms
Exceeds target by 96 ms
RAM
16 kB of 340 kB
ROM
55 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.05 | 0.025 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:15:32 PM

93%
mfe-conv2d-a46
PERFORMANCE
LATENCY
44 ms of 100 ms
RAM
36 kB of 340 kB
ROM
40 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.032 | 0.032 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
dropout - - 0.5

10/27/2022, 5:19:47 PM

93%
mfcc-conv2d-826
PERFORMANCE
LATENCY
95 ms of 100 ms
RAM
16 kB of 340 kB
ROM
55 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.05 | 0.025 | 40

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.5

10/27/2022, 5:43:13 PM

93%
mfe-conv2d-749
PERFORMANCE
LATENCY
159 ms of 100 ms
Exceeds target by 59 ms
RAM
18 kB of 340 kB
ROM
56 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.05 | 0.05 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:40:24 PM

92%
mfe-conv2d-41b
PERFORMANCE
LATENCY
174 ms of 100 ms
Exceeds target by 74 ms
RAM
23 kB of 340 kB
ROM
40 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.032 | 0.032 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
dropout - - 0.25

10/27/2022, 5:48:32 PM

92%
mfcc-conv2d-5e3
PERFORMANCE
LATENCY
677 ms of 100 ms
Exceeds target by 577 ms
RAM
32 kB of 340 kB
ROM
129 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.032 | 0.032 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
conv2d 128 3 -
dropout - - 0.25

10/27/2022, 5:26:27 PM

92%
mfe-conv2d-135
PERFORMANCE
LATENCY
201 ms of 100 ms
Exceeds target by 101 ms
RAM
35 kB of 340 kB
ROM
44 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.02 | 0.02 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
dropout - - 0.5

10/27/2022, 5:51:05 PM

92%
mfcc-conv2d-4a9
PERFORMANCE
LATENCY
136 ms of 100 ms
Exceeds target by 36 ms
RAM
33 kB of 340 kB
ROM
55 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.05 | 0.025 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:42:23 PM

92%
mfcc-conv2d-3b8
PERFORMANCE
LATENCY
485 ms of 100 ms
Exceeds target by 385 ms
RAM
45 kB of 340 kB
ROM
56 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.02 | 0.01 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 8 3 -
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:26:51 PM

92%
mfcc-conv2d-f94
PERFORMANCE
LATENCY
312 ms of 100 ms
Exceeds target by 212 ms
RAM
35 kB of 340 kB
ROM
55 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFCC

0.05 | 0.025 | 40

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
conv2d 64 3 -
dropout - - 0.25

10/27/2022, 5:30:15 PM

92%
mfe-conv2d-785
PERFORMANCE
LATENCY
106 ms of 100 ms
Exceeds target by 6 ms
RAM
28 kB of 340 kB
ROM
41 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.05 | 0.025 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
Data augmentation
conv2d 16 3 -
conv2d 32 3 -
dropout - - 0.5

10/27/2022, 5:32:03 PM

91%
mfe-conv2d-394
PERFORMANCE
LATENCY
202 ms of 100 ms
Exceeds target by 102 ms
RAM
43 kB of 340 kB
ROM
41 kB of 1024 kB
DSP NN Unused
TIME-SERIES INPUT

1000 ms |
1000 ms |
Enabled

MFE

0.05 | 0.025 | 32

ACCURACY
CLASSIFICATION (KERAS)

0.005 | 100

Type Filters Kernel Rate
conv2d 16 3 -
conv2d 32 3 -
dropout - - 0.25

10/27/2022, 5:19:55 PM