Skip to content

Latest commit

 

History

History
58 lines (49 loc) · 2.92 KB

README.md

File metadata and controls

58 lines (49 loc) · 2.92 KB
Raw

Emotion Recognition ER

Emotion recognition is the process of identifying human emotions using AI. This project seeks to recognise emotions from speech clips (audio + video). Generally, the technology works best at this task if it uses multiple modalities, for this reason we implemented a two-stream model to analyze facial expressions from video and voice tone from audio. These tasks are called Speech Emotion Recognition (SER) and Facial Emotion Recognition (FER) respectively.

In this project we trained the models on speech clips of RAVDESS dataset which contains 8 emotion classes: neutral, calm, happy, sad, angry, fearful, surprise and disgust (7 used).

More details about the processing and architecture in Project_Slides.pdf. Dimostrative video and deployed model in the DEMO folder.

Screenshot 2023-03-02 170340

Repo structure

Emotion-Recognition_SER-FER_RAVDESS
├───Datasets
│   ├───RAVDESS
│   ├───RAVDESS_audio
│   ├───RAVDESS_frames
│   ├───RAVDESS_frames_black
│   └───RAVDESS_frames_face_BW
├───DEMO
│   ├───Examples
│   ├───ER DEMO.mp4
│   └───ER_FullClip_DEMO.ipynb
├───Models
│   ├───Audio Stream
│   └───Video Stream
├───Other
│   └───haarcascade_frontalface_default.xml
├───Plots
├───StreamAudio_1D.ipynb
├───StreamAudio_2D.ipynb
├───FullClip_Test.ipynb
├───StreamVideo_FaceOnly.ipynb
├───StreamVideo_FramesExtraction.ipynb
├───StreamVideo_FullFrame.ipynb
├───StreamVideo_Test.ipynb
├───Project_Slides.pdf
├───README.md
├───LICENSE.md
└───requirements.txt

Execution schema

To classify emotions (using our trained model):

  1. Copy your clips in DEMO/Examples
  2. Run ER_FullClip_DEMO.ipynb in DEMO folder

To replicate this project (training and inference):

  1. Download the speech clips of RAVDESS dataset and save it in Datasets/RAVDESS folder
  2. Train video and audio models
    1. Video Stream: extract frames with StreamVideo_FramesExtraction.ipynb (multiple type of frames are generated -> best are "224x224 only faces BW"), train model with StreamVideo_FullFrame.ipynb and StreamVideo_FaceOnly.ipynb (depending on the frames generated) and test the results with StreamVideo_Test.ipynb
    2. Audio Stream: use StreamAudio_1D.ipynb and StreamAudio_2D.ipynb to train models (2D works better)
  3. Use FullClip_Test.ipynb to assess global performance
  4. Use ER_FullClip_DEMO.ipynb in DEMO folder to classify videos with the trained models.

DEMO

ER.DEMO.mp4