SELMA: A Speech-Enabled Language Model for Virtual Assistant Interactions

TL;DR

SELMA is a unified speech-enabled language model that integrates audio and text inputs to improve virtual assistant task performance, simplifying the pipeline and achieving significant accuracy improvements across multiple tasks.

Contribution

It introduces a multi-task end-to-end model with parameter-efficient training and a feature pooling strategy for virtual assistant interactions.

Findings

64% EER reduction on Voice Trigger detection

22% EER reduction on Device-Directed Speech Detection

Near-baseline word error rates on ASR

Abstract

In this work, we present and evaluate SELMA, a Speech-Enabled Language Model for virtual Assistant interactions that integrates audio and text as inputs to a Large Language Model (LLM). SELMA is designed to handle three primary and two auxiliary tasks related to interactions with virtual assistants simultaneously within a single end-to-end model. We employ low-rank adaptation modules for parameter-efficient training of both the audio encoder and the LLM. Additionally, we implement a feature pooling strategy enabling the system to recognize global patterns and improve accuracy on tasks less reliant on individual sequence elements. Experimental results on Voice Trigger (VT) detection, Device-Directed Speech Detection (DDSD), and Automatic Speech Recognition (ASR), demonstrate that our approach both simplifies the typical input processing pipeline of virtual assistants significantly and also improves performance compared to dedicated models for each individual task. SELMA yields relative Equal-Error Rate improvements of 64% on the VT detection task, and 22% on DDSD, while also achieving word error rates close to the baseline.