Statement of Work (SOW)

Project Title:

Assessing the feasibility and usability of a speech-based AI for monitoring team functioning and supporting psychological screening in Mars Exploration Analog (MEA)

Sponsoring Organization:

NASA Human Research Program (HRP), Human Factors and Behavioral Performance

Element

Principal Investigator (PI):

Theodora Chaspari, CU Boulder

Period of Performance:

10/1/2026 – 09/30/2028

1. Background The study is motivated by NASA’s critical need to improve how team functioning is monitored during long‑duration, isolated, and confined space missions. Current reliance on self‑report tools limits detection of psychosocial risks because these methods are subjective, influenced by recall bias, and lack the temporal sensitivity required for early identification of team‑level issues. Speech offers a minimally intrusive, continuous behavioral signal that reflects emotional, cognitive, and social processes, making it a promising alternative for objective monitoring. Prior research has shown that acoustic and linguistic patterns can reliably predict team cohesion, conflict, and performance, suggesting strong potential for operational use. By developing and validating a speech‑based AI system tailored to the Mars

Exploration Analog (MEA), the study aims to provide NASA psychologists with interpretable, data‑driven indicators that can enhance early detection of team‑functioning risks and strengthen behavioral health support.

2. Objectives The overarching objective of the study is to test the feasibility and usability of a speech based artificial intelligence system in the Mars Exploration Analog (MEA) to generate objective, interpretable indicators of team functioning that can support assessment of team functioning during Private Psychological Conferences (PPCs).

Specifically, the project seeks to:

• Adapt and extend existing speech based AI models, previously developed using

HERA mission data, to the MEA environment.

• Evaluate whether acoustic and linguistic markers (tone of voice, prosody, lexical indicators, pronoun use, emotion/cognition-related language) can serve as minimally intrusive predictors of team functioning and performance.

• Develop interpretable AI-supported indicators and visualizations appropriate for operational use by psychologists.

• Integrate and evaluate the system during MEA mission operations, assessing system stability, usability, and operational value.

3. Scope of Work Aim 1: Adapt and Extend Speech Based AI Models for MEA

• Assess MEA audio environment (noise levels, intelligibility, reverberation, device variability).

• Preprocess audio streams using automatic speech recognition (ASR), voice activity detection (VAD), and speaker diarization.

• Extract acoustic and linguistic features such as:

o Fundamental frequency, jitter, shimmer o MFCCs and other spectral measures o Loudness o Pronoun use and linguistic markers of cognitive, affective, and social processes

• Train machine learning models (static and temporal models such as LSTM, autoencoders) to predict team functioning outcomes.

• Compare general models vs. team specific models; evaluate performance, running time, and feasibility for operational cadence.

Aim 2: Design AI Indicators for Supporting Private Psychological Conferences

(PPCs)

• Engage operational psychologists to identify:

o Output types (individual, team-level, biweekly summaries, mission-phase summaries) o Desired granularity of indicators o Preferred visualization formats (dashboards, alerts, temporal trends) o Interpretability and usability needs

• Develop mock system outputs for early-stage iteration.

• Refine indicators, visualizations, and reporting formats through structured feedback sessions.

• Adjust model output content, granularity, and visualization strategies based on psychologist preferences.

Aim 3: Integrate and Evaluate the AI System During the MEA Mission

• Deploy the system in the MEA environment using real-time or near–real-time audio data feeds.

• Continuously evaluate:

o Audio acquisition performance o Preprocessing reliability o Transcription quality o Feature extraction stability o Model drift or instability

• Monitor model output alignment with independent team functioning indicators (self reports, task performance).

• Conduct structured usability assessments with PPC psychologists.

• Iterate system refinement based on:

o Technical performance o Expert user feedback o Operational feasibility

4. Deliverables

• The proposal outlines the following major deliverables:

o Technical and Research Report

▪ Documentation of procedures, model performance, findings, and recommendations related to speech based indicators of team functioning.

▪ Includes analysis results, validation outcomes, and operational integration summaries.

• Speech Based AI System Adapted for MEA o Fully functional, interpretable system capable of generating PPC-facing outputs and includes:

▪ Preprocessing pipeline

▪ Machine learning models

▪ Visualization and reporting components

• Operational Playbook o Guidelines for deploying and using speech-based AI tools in long-duration analog environments and future missions.

o Includes recommended workflows, indicator interpretation guidance, and operational integration processes.

• Curated Datasets and Machine Learning Models o Acoustic and linguistic feature datasets

Q2 FY28

• Finalize Iterative investigation of system outputs and visualizations with operational psychologists

• Operational integration of the system to the MEA

• Deployment and evaluations of the system in the MEA

• Attend and present preliminary data at the HRP Investigator’s Workshop

• Quarterly status call with HFBP

Q3 FY28

• Operational integration of the system to the MEA

• Deployment and evaluations of the system in the MEAQuarterly status call with

HFBP and Interim data presentation and outcomes

Q4 FY28

• Operational integration of the system to the MEA

• Deployment and evaluations of the system in the MEA

• Release software

• Presentations and publications

• Complete Data submission to LSDA

• Final report and presentation to HFBP