Multimodal Large Language Models with Fusion Low Rank Adaptation for Device Directed Speech Detection

TL;DR

This paper introduces FLoRA, a low-rank adaptation method enabling large language models to incorporate multimodal data efficiently, significantly improving device-directed speech detection performance while reducing tuning complexity and maintaining scalability.

Contribution

The paper presents FLoRA, a novel low-rank adaptation technique that allows pre-trained LLMs to effectively integrate new modalities with fewer parameters and enhanced robustness.

Findings

22% relative reduction in EER over text-only models

FLoRA achieves performance parity with full fine-tuning while tuning fewer parameters

Robustness to missing data with 20% lower EER and 56% lower false accept rate

Abstract

Although Large Language Models (LLMs) have shown promise for human-like conversations, they are primarily pre-trained on text data. Incorporating audio or video improves performance, but collecting large-scale multimodal data and pre-training multimodal LLMs is challenging. To this end, we propose a Fusion Low Rank Adaptation (FLoRA) technique that efficiently adapts a pre-trained unimodal LLM to consume new, previously unseen modalities via low rank adaptation. For device-directed speech detection, using FLoRA, the multimodal LLM achieves 22% relative reduction in equal error rate (EER) over the text-only approach and attains performance parity with its full fine-tuning (FFT) counterpart while needing to tune only a fraction of its parameters. Furthermore, with the newly introduced adapter dropout, FLoRA is robust to missing data, improving over FFT by 20% lower EER and 56% lower false accept rate. The proposed approach scales well for model sizes from 16M to 3B parameters.