Emotion Recognition using Speech using Librosa library and MLPClassifier

Overview:

The project deals with understanding emotional state of a human by extracting/detecting the features conveyed through their speech/voice.

Introduction:

Emotions plays an important role in daily interpersonal human interactions. This is essential for our rational as well as intelligent decisions.

Also, detecting emotions is one of the most important marketing strategies in today's world as most of the daily decisions of the humans are driven by their emotional state at that moment.

As a result, this type of application has much potential in the world that would benefit companies and start-ups and also even provide safety to consumers.

Components

Traditional Speech Emotion Recognition system undergoes the following phases to map speech signals into emotions:

1. Speech:

It is given as input (from which emotions are to be extracted)

2. Pre-processing:

The removal of unwanted noise signals from the speech to extract high frequency speech signals.

3. Feature Extraction:

Extarcting the desirable features from the pre-processed speech to identify the human emotions.

4. Classification:

Now, after extracting the features we have to map the features to the corresponding emotions using the classifiers. Some of the traditional classifier are: GMM (Gaussian Mixture Model), ANN (Artificial Neural Network), SVM (Support Vector Machine), etc. For more accurate predictions Deep Learning Models are preferred namely: CNN (Convolutional Neural Networks), etc.

5. Emotion:

It is the output which we want to obtain after training our raw data (i.e., in the form of "utterance") over a suitable classifier.

Technology Used:

- Python

Python Libraries Used:

• Librosa: It is used for analyzing audio and music
• scikit-learn: It is a very robust library in ML and statistical modelling applications including classifications, regression, clustering, etc.
• Numpy: It is used when working with numerical data (specifically; single or multi-dimensional arrays)
• soundfile: It is used for reading and writing sound/audio files

Dataset:

RAVDESS (Ryerson Audio & Visual Database of Emotional Speech and Song)

• The RAVDESS is a multi-modal database of emotional speech and song. The database is gender balanced consisting of 24 professional actor, vocalizing lexically matched statements in a neutral North American accent.
• Speech emotions includes: calm, happy, sad, angry, fearful, surprise, and disgust expressions.
• Each expression is produced at 2 levels of emotional intensity (normal and strong) with an additional neutral expression.
• Speech audio-only files are of 16bit, 48kHz .wav format. Full dataset is of 24.8 GB.
• Kaggle's RAVDESS dataset contains 1440 file: 60 trails per actor x 24 actors = 1440
• Kaggle's audio and speech dataset