PolySpeech: Exploring Unified Multitask Speech Models for Competitiveness with Single-task Models

TL;DR

PolySpeech is a unified multitask speech model that supports recognition, synthesis, and classification, demonstrating competitive performance and benefits over single-task models through joint optimization.

Contribution

It introduces a novel multitask speech model with semantic speech embedding and speech reconstruction, enabling efficient high-quality speech generation and task performance.

Findings

Multitask optimization achieves performance comparable to single-task models.

PolySpeech effectively supports multiple speech tasks within a unified framework.

Joint training benefits specific tasks more than others.

Abstract

Recently, there have been attempts to integrate various speech processing tasks into a unified model. However, few previous works directly demonstrated that joint optimization of diverse tasks in multitask speech models has positive influence on the performance of individual tasks. In this paper we present a multitask speech model -- PolySpeech, which supports speech recognition, speech synthesis, and two speech classification tasks. PolySpeech takes multi-modal language model as its core structure and uses semantic representations as speech inputs. We introduce semantic speech embedding tokenization and speech reconstruction methods to PolySpeech, enabling efficient generation of high-quality speech for any given speaker. PolySpeech shows competitiveness across various tasks compared to single-task models. In our experiments, multitask optimization achieves performance comparable to single-task optimization and is especially beneficial for specific tasks.