Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages

TL;DR

The paper presents USM, a universal speech recognition model trained on 12 million hours of multilingual data, achieving state-of-the-art results across over 100 languages with less labeled data than previous models.

Contribution

It introduces a scalable multilingual pre-training approach with novel techniques like random-projection quantization and modality matching for broad language coverage.

Findings

USM achieves comparable or better performance than Whisper on many languages.

The model performs well on both in-domain and out-of-domain speech recognition tasks.

Pre-training on large unlabeled datasets enables effective multilingual ASR.

Abstract

We introduce the Universal Speech Model (USM), a single large model that performs automatic speech recognition (ASR) across 100+ languages. This is achieved by pre-training the encoder of the model on a large unlabeled multilingual dataset of 12 million (M) hours spanning over 300 languages, and fine-tuning on a smaller labeled dataset. We use multilingual pre-training with random-projection quantization and speech-text modality matching to achieve state-of-the-art performance on downstream multilingual ASR and speech-to-text translation tasks. We also demonstrate that despite using a labeled training set 1/7-th the size of that used for the Whisper model, our model exhibits comparable or better performance on both in-domain and out-of-domain speech recognition tasks across many languages.