SSR: Alignment-Aware Modality Connector for Speech Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tan, Weiting, Inaguma, Hirofumi, Dong, Ning, Tomasello, Paden, Ma, Xutai
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910949001134080
author Tan, Weiting
Inaguma, Hirofumi
Dong, Ning
Tomasello, Paden
Ma, Xutai
author_facet Tan, Weiting
Inaguma, Hirofumi
Dong, Ning
Tomasello, Paden
Ma, Xutai
contents Fusing speech into pre-trained language model (SpeechLM) usually suffers from inefficient encoding of long-form speech and catastrophic forgetting of pre-trained text modality. We propose SSR-Connector (Segmented Speech Representation Connector) for better modality fusion. Leveraging speech-text alignments, our approach segments and compresses speech features to match the granularity of text embeddings. Additionally, we introduce a two-stage training pipeline that includes the distillation and fine-tuning phases to mitigate catastrophic forgetting. SSR-Connector outperforms existing mechanism for speech-text modality fusion, consistently achieving better speech understanding (e.g., +10 accuracy on StoryCloze and +20 on Speech-MMLU) while preserving pre-trained text ability.
format Preprint
id arxiv_https___arxiv_org_abs_2410_00168
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SSR: Alignment-Aware Modality Connector for Speech Language Models
Tan, Weiting
Inaguma, Hirofumi
Dong, Ning
Tomasello, Paden
Ma, Xutai
Computation and Language
Sound
Audio and Speech Processing
Fusing speech into pre-trained language model (SpeechLM) usually suffers from inefficient encoding of long-form speech and catastrophic forgetting of pre-trained text modality. We propose SSR-Connector (Segmented Speech Representation Connector) for better modality fusion. Leveraging speech-text alignments, our approach segments and compresses speech features to match the granularity of text embeddings. Additionally, we introduce a two-stage training pipeline that includes the distillation and fine-tuning phases to mitigate catastrophic forgetting. SSR-Connector outperforms existing mechanism for speech-text modality fusion, consistently achieving better speech understanding (e.g., +10 accuracy on StoryCloze and +20 on Speech-MMLU) while preserving pre-trained text ability.
title SSR: Alignment-Aware Modality Connector for Speech Language Models
topic Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2410.00168