Skip to content

This repo is a streamlit app that records audio in browser, and helps to find the silimlarity between 2 or 3 classes of sound.

Notifications You must be signed in to change notification settings

satyakamacodes/sound_clustering_visualization_verOne

Repository files navigation

Machine States Identification from Audio

The objective of this project is to visualise multiple states of machines using the audio emitted by each of the states. Here, by states, we loosely refer to the conditions of the machine such as GOOD vs. MALFUNCTIONING; BRAND NEW vs. WORKS DECENTLY vs.DILIAPIDATED etc.

App available at:

APP(Streamlit)

Possible Use Cases:

  • In an industrail setting where you quickly need to compare if one machine is functioning as good as another.

Features:

  • In browser recording of sound
  • Multiple dimensionality reduction techniques to better visualise overlap or seperation between the states.

Best suited for:

  • Non human sounds, speechs etc and primarily machine sounds.

Folder and file structure

  1. main.pycontains the streamlit app.
  2. source contains the function definitions.
  3. Note: there are two txt files in the recordings_three_class and recordings_two_class - do not remove them.

Audio feature extraction presently is through

  1. MFE
  2. Proposed - SPectral Grating

Dimensionality reduction

  1. Presently through: PCA and t-SNE
  2. Proposed: UNET

Logical flow

title
    |
    |---how many types of machine states
        |
        |---2
        |   |
        |   |---NORMAL vs FAULTY
        |   |---NORMAL, FAULTY, EXTREMELY_FAULTY  ------|
        |                                               | 
        |---specify sampling rate                       |
        |---specify duration                            |
                    |                                   |
                    |                                   |
                    |-----------------------------------|
                                                        |
                                                        |--- Saving files of each class
                                                             |----Calculating
                                                                   |----MFCC
                                                                   |----MFE
                                                                   |----Spectral Grating
                                                                        |
                                                                        |---Dimensionality reduction using:
                                                                                |--- PCA
                                                                                |--- tSNE
                                                                                |----UMAP
                                                                                        |
                                                                                        |-- Visualization
                                                                                                |---- 2D Scatter
                                                                                                |---- 3D Scatter
                                                                                
                                                        |

            

Screen shots of the app

Screenshot One Screenshot Two

About

This repo is a streamlit app that records audio in browser, and helps to find the silimlarity between 2 or 3 classes of sound.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages