knnkeywordspotting's Introduction

Project Introduction

A-Lightweight-Model-for-Keyword-Spotting by Webersan

The project is in the development of continuous improvement.

Environment

Python 3.10.3

LIBs				   VERSION
numpy                  1.21.6
scipy                  1.8.0
PyAudio                0.2.11
python-speech-features 0.6
matplotlib             3.5.1	(not necessary)
scikit-learn           1.0.2	(not necessary)

A Way to Test the Model

python main.py

Modify test time duration

Change N to modify time duration, about 2.5 secs per 100 chunks.

N == 300 is suggested.

for _ in range(N):	# main.py line 45

Result

While the program is running, you are supposed to speak to the microphone on your computer. The console displays a string containing the most probable phoneme.

Listening...
c:\PythonProjects\DSP Project\main.py:50: DeprecationWarning:
The binary mode of fromstring is deprecated, as it behaves surprisingly on unicode inputs.
Use frombuffer instead data_int = np.fromstring(data_bytes, dtype=np.int16)

NNNNNNNNNNNNNNNNNNNNNNNNNssssssssoooooooorioNNotoooo
NNNNriririririririNNNNNNoooooooooooooooooooooooNNNNN
tttNtNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN
NNNNNNNNNtttttNNttriNNsssssssoooooooNNNririririririri
riririririritNNNNooooooNNNNNttNtNNNNNririririNriNNNN
NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNriNNNNNNNNNNNNNNNNNN
NNNNNNNNNNNNNNN

Time cost: 7.512481212615967 s
Listening Complete.

Up to now, the model supports these phoneme:

's'

'o'

'ri' (or pricisely 'i')

't'

N (nothing)

This model recognizes vowels better than consonants.

Recommend Projects