This project's objective is to recognize speech without triggers. Trigger means a command indicating the start of an instruction such as "OK google" or "Hey siri".
Follow the steps below to speech recognition.
- Record instruction voice
- Convert recorded audio data to mfcc (Mel-Frequency Cepstrum Coefficients)
- Infer the mfcc with CNN
The registered voice commands are as follows. (Can be added other commands)
- 「電気をつけて」(Turn on the light)
- 「電気を消して」(Turn off the light)
- 「エアコンつけて」(Turn on the air conditioner)
- 「エアコン消して」(Turn off the air conditioner)
- 「カーテン開けて」(Open the curtain)
- 「カーテン閉めて」(Close the curtain)
- 「テレビをつけて」(Turn on the TV)
- 「テレビを消して」(Turn off the TV)
- 「扇風機をつけて」(Turn on the fan)
- 「扇風機を消して」(Turn off the fan)
- Python 3.5.2
- Tensorflow 2.0.0a0
- Keras 2.2.4
$ git clone https://github.com/Crtv-info/Speech_Reicognition
$ cd Speech_Recognition
- Run a program to record audio
$ python3 detection.py
- Run a program to infer
$ python3 api.py