ICML ExVo 2022 : ICML Expressive Vocalization (ExVo) Workshop and Competition
Call For Papers
We are delighted to announce that the first annual ICML Expressive Vocalization (ExVo) Workshop and Competition will be open for registration on April 1st 2022. Organized by leading multidisciplinary researchers in emotion science and AI research and co-sponsored by Hume AI and Mila, in this first iteration of ExVo, we explore the machine learning problem of understanding and generating vocal bursts—a wide range of expressive non-verbal vocalizations such as laughs, sighs, grunts, cries, and screams (to name but a few).
Participants of the ExVo Challenge will be presented with three tasks that utilize a newly introduced, large-scale dataset with 60k recordings from 1.7k speakers. The dataset and three tasks draw attention to new innovations in emotion science and capture 10 dimensions of emotion reliably perceived in distinct vocal bursts: Awe, Excitement, Amusement, Awkwardness, Fear, Horror, Distress, Triumph, Sadness and Surprise.
The three tasks include:
The Multi-task High-Dimensional Emotion, Age & Country Task (ExVo Multi-task). Participants will train models to predict the average intensity of each of 10 emotions perceived in vocal bursts, as well as the speaker's age and country, using multitask learning.
The Generative Emotional Vocal Burst Task (ExVo Generate). Participants will use generative models to produce vocal bursts that are associated with 10 distinct emotions. Each team will submit five machine-generated vocalizations that separately convey each emotion—“awe,” “fear,” and more—with maximal intensity and fidelity. The ExVo organization team will provide an automated method for scoring generated samples, but the final evaluation will be performed using human survey ratings.
The Few-Shot Emotion Recognition task (ExVo Few-Shot). Participants will predict 10 emotions associated with each speaker’s vocal bursts using multi-output regression with two-shot speaker personalization. Participants will be provided with at least 2 labeled samples per speaker in all splits (train, validation, test). The speaker IDs and the 2 labeled samples per speaker in the test set will be withheld until a week before the deadline for final evaluation of ExVo Few-Shot models.
The ExVo Workshop will also be accepting contributions on other related topics:
Detecting and Understanding Vocal Emotional Behavior
Multi-Task Learning in Affective Computing
Generating Nonverbal Vocalizations or Speech Prosody
Personalized Machine Learning for Affective Computing
Other topics related to Affective Verbal and Nonverbal Vocalization
See the following website for more information, rules and deadlines: http://www.competitions.hume.ai
Attached you will find a CfP which can be shared with interested colleagues, and the general deadlines are as follows:
Challenge start (data release): April 1st
Baselines and White paper out: April 8th
Submission for participants and papers: June 3rd
Notification of Acceptance: June 12th
Camera-Ready, Posters/Slides: June 17th
Workshop TBD: July 22/23
We look forward to hearing from interested parties! Please get in touch with any questions!
ExVo Workshop & Competition organizing team