BriefGPT.xyz
Nov, 2022
一种弱监督的流式多语言语音模型，具有真正的零-shot能力
A Weakly-Supervised Streaming Multilingual Speech Model with Truly Zero-Shot Capability
HTML
PDF
Jian Xue, Peidong Wang, Jinyu Li, Eric Sun
TL;DR
本文介绍了建立流式多语言语音模型 (SM2) 的工作，基于 Transformer Transducer，使用弱监督数据通过机器翻译服务训练模型，拥有较强的流式能力和真正的零-shot 能力，并取得了非常好的翻译质量。
Abstract
In this paper, we introduce our work of building a
streaming multilingual speech model
(SM2), which can transcribe or translate multiple spoken languages into texts of the target language. The backbone of SM2 is
transfo
→