SBIR Phase II: Automated Emotional Distress Severity Classification for Children and Adolescents Using Speech Emotion Recognition and AI

这项小企业创新研究（SBIR）第二阶段项目的广泛影响和商业潜力将利用新兴的语音情感识别和人工智能（AI）科学，为低收入社区的儿童和青少年提供个性化的行为医疗保健服务。迫切需要一个平台来客观地衡量精神障碍，并使多个利益攸关方能够合作，提供负担得起的优质精神卫生保健。该项目开发了一种谈话治疗软件，可以从60秒的语音样本中提取语音生物标志物，以测量儿童和青少年情绪困扰的严重程度。该平台基于语音的算法增强了治疗师的临床能力，提高了社区精神卫生提供者解决质量和获取问题的能力。该项目开发了一种基于语音的生物标记算法，经过训练可以理解行为和情感倾向，并预测未来的行为，以确定儿童的声音是否偏离了与年龄相适应的语言和语音模式。该公司正在开发一个专有的临床语音样本数据库和存储库，代表边缘化社区（非洲裔美国人、拉丁裔和农村社区的高加索人），其数量和准确性将超过业内同行。如今，该公司的数据库包含了一套多样化的、基础的、规模庞大的语音样本，这对于开发一种更准确的算法来检测和预测情绪障碍的严重程度至关重要。第二阶段的拟议研究建立在第一阶段取得的进展的基础上，使用语音情感识别和机器学习（ML）来识别创伤和压力的可测量生物标志物。该方法将创伤、压力和声音类型联系起来。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact and commercial potential of this Small Business Innovation Research (SBIR) Phase II project will leverage the emerging science of Speech Emotion Recognition and artificial intelligence (AI) to transform the delivery of personalized behavioral healthcare services for children and adolescents from low-income communities. A strong need exists for a platform to measure mental disorders objectively and enable collaboration between multiple stakeholders for affordable quality mental healthcare. This project advances a talk therapy software that extracts voice biomarkers from 60-second speech samples to measure the severity of emotional distress for children and adolescents. The platform’s voice-based algorithm augments therapists' clinical capabilities – increasing the capacity of community mental health providers to address both quality and the issue of access. The proposed project develops a voice-based biomarker algorithm trained to understand behavioral and emotional tendencies and anticipate future behaviors to determine if a child’s vocal utterances deviate from age-appropriate linguistic and speech patterns. The company is developing a proprietary clinical voice sample database and repository representing marginalized communities (African American, Latino, and Caucasian within rural communities) that will exceed in both volume and accuracy those of its industry peers. Today, the company’s database consists of a diverse, foundational set and size of voice samples essential for developing a more accurate algorithm(s) to detect and predict emotional disorder severity. The proposed research within Phase II builds on the progress achieved in Phase I, using speech emotion recognition and Machine Learning (ML) to identify measurable biomarkers for trauma and stress. The methodology links trauma, stress, and voice types.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.