This is a basic speech recognition example. For more information, see the tutorial at https://www.tensorflow.org/tutorials/sequences/audio_recognition
tensorflow==1.15.0 Optional: speech command database (If you want to train again, https://www.tensorflow.org/datasets/catalog/speech_commands)
In run.sh
file:
datadir=new_data # Directory where the speech data is stored
wanted_words_list='busagent,hellobus,okagent,okbus'
These two lines indicate directory and wanted words. The directory has
new_data/
|-- busagent
|-- hellobus
|-- okagent
|-- okbus
|-- _background_noise_
You can record data and put inside these folders. background_noise refers to the audio other than wakeword speech.
use run.sh
- Load pretrained model
pretrained_pb2npz.py
- Run the transfer learning
transfer.py
- Convert back to protobuffer
freeze.py
- Do quick test
- Convert to tflite
To bulid the container
docker build . -t wakeup
Running training and testing file,i.e, run.sh
within the container
- Run the container with
bash
entrypoint:
$ docker run -it wakeup bash
- You will be inside root, then run
run.sh
.