Emozionalmente: a crowdsourced Italian speech emotional corpus

This repository contains Emozionalmente: an extensive Italian speech emotional corpus. The dataset consists of 6902 labeled samples acted out by 431 amateur actors while verbalizing 18 different sentences expressing the Big Six emotions (anger, disgust, fear, joy, sadness, surprise) plus neutrality. Labels represent the emotional communicative intention of the actors (i.e., the seven emotional states). 

Recordings were generally obtained with non-professional equipment. They are .wav files, mono-channel, and have a sample size of 16 bits and a sample rate of 16000 Hz. Each audio recording lasts 3.81 seconds (SD = 0.99 seconds). 

We validated the emotional content of the clips by asking 829 humans (5 evaluations per audio) to guess the emotion contained in each recording. Humans obtained a general accuracy of 66% (comparable with previous literature in the field).

Beyond the audio data, in the repo, you can find three .csv files describing the demographics of the actors and the evaluators and the emotions they expressed and recognized for each audio. 

Soon we will publish a paper with more information about the dataset, including data acquisition methodology, participant demographics, and data validation. Also, to help your research in Speech Emotion Recognition, we share the code for training and testing an emotional classifier on the Emozionalmente dataset (81% accuracy).

If you use this dataset, please cite "Emozionalmente: a crowd-sourced Italian speech emotional corpus" by Fabio Catania.