Speech emotion recognition

Abstract Emotion recognition or affect detection from speech is an old and challenging problem in the field of artificial [login to view URL] significant research works have been done on emotion Recognition by using different approaches as it. can be useful in the field of medical science , robotics engineering, call center application etc. Human can easily recognize emotion of speaker but the question is machine can? Flow of the System Identification of emotion can be done by extracting the features or different characteristics from the speech and a training is needed for a large number of speech database to make the system accurate. The steps towards building of an emotion recognition system are, an emotional speech corpora is selected or implemented then emotion specific features are extracted from those speeches and finally a classification model is used to recognize the Emotions. Tentative Project Module Names with Description: 1 ) Data Processing Gather the libraries and test their functions such as import tensorflow import keras import pickle #To store our model Load the data into diffrent folders containing angry and neutral audio files. 2) Featurization Before going into pre-processing and data exploration to select our features. Mel scale — deals with human perception of frequency, it is a scale of pitches judged by listeners to be equal distance from each other. Pitch — how high or low a sound is. It depends on frequency, higher pitch is high frequency. Frequency — speed of vibration of sound, measures wave cycles per second. Chroma — Representation for audio where spectrum is projected onto 12 bins representing the 12 distinct semitones (or chroma). Computed by summing the log frequency magnitude spectrum across octaves. Fourier Transforms — used to convert from time domain to frequency domain. Time domain shows how signal changes over time. Expected Outcome : How much of the signal lies within each given frequency band over a range of frequencies 3) Model Training We will begin with a simple model architecture, consisting of three layers - an input layer, a hidden layer and an output layer. All three layers will be of the dense layer type which is a standard layer type that is used in many cases for neural networks. We will also apply a Dropout value of 50% on our first two layers. This will randomly exclude nodes from each update cycle which in turn results in a network that is capable of better generalisation and is less likely to overfit the training data. For compiling our model, we will use the following three parameters: Loss function - we will use categorical_crossentropy. This is the most common choice for classification. A lower score indicates that the model is performing better. Metrics - we will use the accuracy metric which will allow us to view the accuracy score on the validation data when we train the model. Optimizer - here we will use adam which is a generally good optimizer for many use cases. After that We will start with 100 epochs, which is the number of times the model will cycle through the data. The model will improve on each cycle until it reaches a certain point. Lastly, We will store our model in a pickle file, so we don't have to keep training it again and again while testing. 4) Testing: Extract the features, then print the prediction in percentage. 5) Visualization and data Storage. User will be given options to upload a audio file then our algorithm will tell the particular mood with the percentage and will give graph of emotions. This all data will be stored in a database(sqlite3).