TensorFlow Keyword Spotting

Native C/C++. Suitable for embedded devices.

Demo

Default models pre-trained on 0-9 words: zero one two three four five six seven eight nine.

~$ arecord -f S16_LE -c1 -r16000 -d1 test.wav
~$ aplay test.wav
~$ dataset/dataset/google_speech_commands/src/features/build.sh
~$ src/brain/build.sh
~$ alias fe=dataset/dataset/google_speech_commands/bin/fe
~$ fe test.wav | bin/guess models/mlp.tflite
~$ fe test.wav | bin/guess models/cnn.tflite
~$ fe test.wav | bin/guess models/rnn.tflite
~$ fe test.wav | bin/guess models/dcnn.tflite

Training

Jupyter Notebooks MLP | CNN | RNN | DCNN.

Each notebook generates model file. To evaluate model accuracy:

~$ apt install gcc lrzip wget
~$ wget https://github.com/42io/dataset/releases/download/v1.0/0-9up.lrz -O /tmp/0-9up.lrz
~$ lrunzip /tmp/0-9up.lrz -o /tmp/0-9up.data # md5 87fc2460c7b6cd3dcca6807e9de78833
~$ dataset/matrix.sh /tmp/0-9up.data 12 # num outputs

Confusion matrix for pre-trained modeles:

MLP confusion matrix...
zero   0.93 0.00 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.03 0.00 | 603
one    0.00 0.85 0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.05 0.06 0.01 | 575
two    0.03 0.00 0.86 0.02 0.02 0.00 0.00 0.01 0.01 0.00 0.04 0.01 | 564
three  0.00 0.00 0.01 0.90 0.00 0.01 0.01 0.01 0.04 0.01 0.01 0.01 | 548
four   0.00 0.01 0.01 0.00 0.90 0.01 0.00 0.00 0.00 0.00 0.05 0.01 | 605
five   0.00 0.01 0.00 0.01 0.01 0.80 0.01 0.03 0.01 0.03 0.09 0.01 | 607
six    0.00 0.00 0.00 0.00 0.00 0.00 0.96 0.00 0.00 0.00 0.02 0.01 | 462
seven  0.01 0.00 0.03 0.01 0.00 0.00 0.01 0.90 0.00 0.00 0.03 0.01 | 574
eight  0.00 0.00 0.01 0.07 0.00 0.00 0.03 0.00 0.84 0.01 0.03 0.01 | 547
nine   0.00 0.04 0.00 0.01 0.00 0.01 0.00 0.01 0.00 0.86 0.06 0.01 | 596
#unk#  0.02 0.03 0.03 0.05 0.06 0.07 0.02 0.03 0.02 0.07 0.58 0.02 | 730
#pub#  0.00 0.00 0.01 0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.96 | 730
MLP guessed wrong 1029...

CNN confusion matrix...
zero   0.97 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.00 | 603
one    0.00 0.93 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.01 0.05 0.00 | 575
two    0.01 0.00 0.95 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.03 0.00 | 564
three  0.00 0.00 0.00 0.91 0.00 0.00 0.01 0.01 0.01 0.00 0.06 0.00 | 548
four   0.00 0.00 0.00 0.00 0.90 0.00 0.00 0.00 0.00 0.00 0.09 0.00 | 605
five   0.00 0.00 0.00 0.00 0.00 0.93 0.00 0.00 0.01 0.01 0.06 0.00 | 607
six    0.00 0.00 0.00 0.00 0.00 0.00 0.99 0.00 0.00 0.00 0.01 0.00 | 462
seven  0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.97 0.00 0.00 0.02 0.00 | 574
eight  0.00 0.00 0.01 0.01 0.00 0.01 0.01 0.00 0.93 0.00 0.03 0.00 | 547
nine   0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.93 0.06 0.00 | 596
#unk#  0.01 0.01 0.00 0.02 0.02 0.00 0.00 0.00 0.00 0.01 0.92 0.01 | 730
#pub#  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.98 | 730
CNN guessed wrong 427...

RNN confusion matrix...
zero   0.98 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 | 603
one    0.00 0.95 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.01 0.02 0.00 | 575
two    0.00 0.00 0.98 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 | 564
three  0.00 0.00 0.00 0.97 0.00 0.00 0.01 0.00 0.01 0.00 0.01 0.00 | 548
four   0.00 0.00 0.00 0.00 0.97 0.00 0.00 0.00 0.00 0.00 0.02 0.00 | 605
five   0.00 0.00 0.00 0.00 0.01 0.98 0.00 0.00 0.00 0.00 0.01 0.00 | 607
six    0.00 0.00 0.00 0.00 0.00 0.00 0.99 0.00 0.00 0.00 0.00 0.00 | 462
seven  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.98 0.00 0.00 0.01 0.00 | 574
eight  0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.97 0.00 0.01 0.00 | 547
nine   0.00 0.01 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.97 0.02 0.00 | 596
#unk#  0.00 0.01 0.00 0.01 0.02 0.02 0.00 0.00 0.01 0.02 0.91 0.00 | 730
#pub#  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.99 | 730
RNN guessed wrong 220...

DCNN confusion matrix...
zero   0.98 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 | 603
one    0.00 0.98 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.01 0.00 | 575
two    0.01 0.00 0.98 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 | 564
three  0.00 0.00 0.00 0.97 0.00 0.00 0.01 0.00 0.01 0.00 0.00 0.00 | 548
four   0.00 0.00 0.00 0.00 0.98 0.00 0.00 0.00 0.00 0.00 0.01 0.00 | 605
five   0.00 0.00 0.00 0.00 0.00 0.98 0.00 0.00 0.00 0.00 0.01 0.00 | 607
six    0.00 0.00 0.00 0.00 0.00 0.00 0.99 0.00 0.00 0.00 0.00 0.00 | 462
seven  0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 | 574
eight  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.99 0.00 0.01 0.00 | 547
nine   0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.98 0.01 0.00 | 596
#unk#  0.00 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.94 0.00 | 730
#pub#  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 | 730
DCNN guessed wrong 143...

Heap Memory Usage

Some magic numbers to know before stepping into embedded world.

~$ valgrind dataset/dataset/google_speech_commands/bin/fe test.wav # 606,416 bytes allocated
~$ fe test.wav | valgrind bin/guess models/mlp.tflite              # 347,138 bytes allocated
~$ fe test.wav | valgrind bin/guess models/cnn.tflite              # 1,793,114 bytes allocated
~$ fe test.wav | valgrind bin/guess models/rnn.tflite              # 2,442,810 bytes allocated
~$ fe test.wav | valgrind bin/guess models/dcnn.tflite             # 595,958 bytes allocated