This paper presents Voicebox, the most versatile text-conditioned speech generative model at scale. Voicebox is trained on a text-guided speech infilling task, where the goal is to generate masked speech given its surrounding audio and text transcript.
To address these problems and encourage re- search to develop NLU technologies in the privacy policy domain, we introduce the Privacy Policy Language Understanding Evaluation (PLUE) benchmark, to evaluate the privacy policy language understanding across six tasks, including text classification, question answering, semantic parsing, and named-entity recognition.
This paper presents Cocktail HuBERT, a self-supervised learning framework that generalizes to mixture speech using a masked pseudo source separation objective.
Neural network pruning compresses automatic speech recognition (ASR) models effectively. However, in multilingual ASR, language ...
End-to-end multilingual ASR has become more appealing because of several reasons such as simplifying the training and deployment process and positive performance ...
To mitigate undersampling, our approach inflates the uncertainty lower bound and weights each loss component with their uncertainty, effectively compensating severely...
We propose a new neural network architecture for the task of sequential audio tagging. "Sequential audio tagging" means we want to know what types of acoustic events (e.g. dog bark, car engine) occur in an audio recording, and in what order they occur.
Visual speech recognition (VSR), also known as lip reading, is the task of recognizing speech content based on visual lip movements. VSR has a wide range of applications in real-world scenarios such as helping the hearing- impaired perceive human speech and improving automatic speech recognition (ASR) in noisy environments.
In this work, we propose LA-VocE, a new two-stage approach that predicts mel-spectrograms from noisy audiovisual speech via a transformer-based architecture, and then converts...
we build a new dataset comprising a moderate amount of labeled data for 1,107 languages and another dataset of unlabeled speech in 3,809 languages (§3). We leverage ....
