https://github.com/sk-g/speech-recognition-tensorflow-challenge

Different CNN Models for keyword spotting in speech recognition
https://github.com/sk-g/speech-recognition-tensorflow-challenge

alexnet-model cnn deep-learning machine-learning neural-networks speech-recognition tensorflow

Last synced: about 2 months ago
JSON representation

Different CNN Models for keyword spotting in speech recognition

• Host: GitHub
• URL: https://github.com/sk-g/speech-recognition-tensorflow-challenge
• Owner: sk-g
• License: gpl-3.0
• Created: 2017-12-19T20:05:52.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2018-07-11T23:38:08.000Z (about 7 years ago)
• Last Synced: 2025-04-03T18:13:03.842Z (6 months ago)
• Topics: alexnet-model, cnn, deep-learning, machine-learning, neural-networks, speech-recognition, tensorflow
• Language: Python
• Homepage:
• Size: 1.24 GB
• Stars: 10
• Watchers: 0
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: ReadMe
  • License: LICENSE

Awesome Lists containing this project

README

          
### Update 10/07/2018

## MAJOR CHANGES COMING IN SOON, inlcuding pytorch implementation and better structure

Tensorflow Speech Recognition Challenge

https://www.kaggle.com/c/tensorflow-speech-recognition-challenge

Folders : 

	images: audio clips -> spectrogram images 

	im_train: -> images -> resize to 28x28

	results: results in graphs

	papers: some useful papers

	test_pics : ignore (spectrograms of test audio clips)

	Deprecated : old GCP files. Ignore

Files :

	complete.py -> code with two CNN models and adversarial training

	ReadMe -> this

	Some files were used for preprocessing on older data

	but maybe useful for other projects

	ignore these:

		CNN_code_for_resized_data.py

		dataset.py

		downsizing.py <- recursively resize all images in a folder

		ds.py <- tried an iterator

		pp.py <- audio to image conversion. recursively converts all audio clips in a folder to 

				 corresponding spectrograms

		speech_recog.py <- ignore

		GCP-SR.py <-- for local usage in google cloud platform

		

Models:

Shallow CNN: CNN similar to AlexNet. Two fc layers at the end, dropout enabled/disabled.

Deeper CNN:

	wide  : added more layers to the CNN, removed dropout

	wider : increased number of filters

For Results and Talks:

	ML_final.pdf

	ML_talk.pdf