Interspeech 2025 official challenge

Introduction

Suicide is a significant global health problem, especially for young people. The SpeechWellness Challenge aims to detect suicide risk in teenagers through speech analysis. Using both spontaneous and reading speech data from 600 adolescents, including those identified as at risk, participants aim to develop models to identify digital biomarkers associated with suicidal tendencies. The challenge encourages diverse approaches, including signal processing, speech and emotion recognition, and large language models. By using speech data, the challenge aspires to advance the early detection of mental health issues and provide scalable, accessible tools for intervention in adolescent suicide prevention.

The Challenge

The 1st SpeechWellness challenge consists of a suicidal risk detection task, where participants are required to produce a model to predict the label (has suicidal risk or not) for a subject. The training and testing dataset comprises speech recordings from 600 Chinese teenagers aged 10-18 years, with 50% identified as having suicidal risk based on psychological scales. All subjects (speakers) have been anonymized before making this dataset available to the participants, which provides a unique opportunity to apply and refine advanced speech technologies for public health while preserving the privacy of all subjects.

Participants are encouraged to leverage a variety of methodologies, including but not limited to signal processing, self-supervised learning foundation models, speech recognition, and large language models. The goal is to push the boundaries of current technology to create effective tools for early detection and intervention.

The performance will be evaluated by accuracy:

Apart from accuracy, the participants are also encouraged to report other metrics (e.g., F1, AUROC).

How to Participate

To register for the SpeechWellness Challenge, please email speechwellnesschallenge@gmail.com with your contact information, affiliation, and agree to the terms and conditions. All participants must register by January 15, 2025. A confirmation email with further instructions on accessing the dataset will be sent to registered participants.

Participants may withdraw from the challenge at any time by notifying the organizers. Upon withdrawal, they must delete all copies of the dataset and related materials.

Data Access and Usage

Registered participants will be provided access to the training and testing datasets. They must sign a data usage agreement (see below), agreeing to maintain confidentiality and follow data protection protocols. The dataset is to be used solely for challenge purposes, with strict prohibitions on redistribution or any other use. Participants are responsible for securing the data from unauthorized access.

The dataset contains sensitive information from minors aged 10-18, making it essential for all participants to strictly adhere to ethical guidelines and comply with relevant legal regulations. Even if the data has been anonymized, participants MUST refrain from attempting to identify or contact any individuals whose data is included in the dataset. It is crucial to understand that anonymization does not eliminate the ethical responsibility to protect the privacy and dignity of the subjects. Any findings must be reported with the utmost care to ensure that no harm, stigma, or unintended consequences arise from the research. Additionally, participants are expected to prioritize the well-being of the individuals represented in the data by avoiding conclusions or interpretations that may inadvertently lead to their identification or negatively impact their communities. Failure to uphold these standards could result in legal or disciplinary action and could compromise the integrity of the research.

Data License Agreement

Please sign the license and return by email to speechwellnesschallenge@gmail.com 

Important Dates

The following dates are in accordance with Interspeech 2025:

Participants must submit a paper detailing their methods, experiments, and results to the Interspeech 2025 main conference (select subject area “14.15 The 1st SpeechWellness Challenge [Challenge]”), following the Interspeech 2025 submission guidelines. Papers should adhere to the Interspeech 2025 formatting instructions, and represent original, unpublished work.

The submission will undergo the official peer review process for Interspeech 2025. The evaluation will consider both quantitative performance metrics and qualitative aspects of the work, including methodological innovation, reproducibility, presentation, and ethical considerations.

Contact

For any inquiries related to the challenge, please contact, please contact speechwellnesschallenge@gmail.com

Organizers