Qfin Holdings Achieves Major Milestone at ASRU 2025 with Groundbreaking Speech Technology Research

Qfin Holdings Achieves Major Milestone at ASRU 2025

In a remarkable display of innovation, Qfin Holdings' Intelligent Speech Team has successfully unveiled its latest research paper at the prestigious ASRU 2025 conference. The paper, titled "Qieemo Multimodal Emotion Recognition Based on the ASR Backbone," not only garnered acceptance at this global gathering but also cemented Qfin Holdings' stature as a powerhouse in the realm of speech technology.

ASRU (IEEE Workshop on Automatic Speech Recognition and Understanding) is revered as a premier event in the field of audio comprehension. It showcases groundbreaking studies and methodologies that shape the future of speech recognition systems. With the acceptance of its paper, Qfin Holdings joins the elite ranks of global innovators in the realm of speech technology, recognized across the industry's top conferences including ICASSP and InterSpeech.

The core significance of the paper lies in its establishment of a universal theoretical framework, moving beyond task-specific models. By leveraging advanced mathematical modeling techniques, the study introduces a general feature fusion framework that pivots on the ASR model as its fundamental backbone. It thoroughly explores the principal contributions and mechanisms of multi-level features derived from a pre-trained ASR model encoder, extending their relevance to downstream tasks in audio understanding. Moreover, the framework innovatively diverges from traditional methods that generally involve merely enhancing network layers or fine-tuning existing models. Instead, the research delves into the essence of speech representation and the interconnected dynamics of cross-modal operations, thereby offering a fresh theoretical foundation for multimodal emotion recognition.

The revelations in the study also showcase a significant increase in recognition accuracy—over 15% better than conventional techniques. Additional accomplishments include a notable improvement under complex scenarios, achieving a 3.04% relative enhancement over the state-of-the-art single-modal method, MSMSER. These advancements empower intelligent customer service systems with the ability to genuinely comprehend emotions, establishing a new benchmark—dubbed