Salah Karout

Bethan Thomas, Samuel Kessler, Salah Karout

Self-supervised learning (SSL) is a powerful tool that allows learning of underlying representations from unlabeled data. Transformer based models such as wav2vec 2.0 and HuBERT are leading the field in the speech domain. Generally these models are fine-tuned on a small amount of labeled data for a downstream task such as Automatic Speech Recognition (ASR). This involves re-training the majority of the model for each task. Adapters are small lightweight modules which are commonly used in Natural Language Processing (NLP) to adapt pre-trained models to new tasks. In this paper we propose applying adapters to wav2vec 2.0 to reduce the number of parameters required for downstream ASR tasks, and increase scalability of the model to multiple tasks or languages. Using adapters we can perform ASR while training fewer than 10% of parameters per task compared to full fine-tuning with little degradation of performance. Ablations show that applying adapters into just the top few layers of the pre-trained network gives similar performance to full transfer, supporting the theory that higher pre-trained layers encode more phonemic information, and further optimizing efficiency.

Duo Wang, Salah Karout

Multi-Modal Self-Supervised Learning from videos has been shown to improve model's performance on various downstream tasks. However, such Self-Supervised pre-training requires large batch sizes and a large amount of computation resources due to the noise present in the uncurated data. This is partly due to the fact that the prevalent training scheme is trained on coarse-grained setting, in which vectors representing the whole video clips or natural language sentences are used for computing similarity. Such scheme makes training noisy as part of the video clips can be totally not correlated with the other-modality input such as text description. In this paper, we propose a fine-grained multi-modal self-supervised training scheme that computes the similarity between embeddings at finer-scale (such as individual feature map embeddings and embeddings of phrases), and uses attention mechanisms to reduce noisy pairs' weighting in the loss function. We show that with the proposed pre-training scheme, we can train smaller models, with smaller batch-size and much less computational resources to achieve downstream tasks performances comparable to State-Of-The-Art, for tasks including action recognition and text-image retrievals.

Samuel Kessler, Bethan Thomas, Salah Karout

We present a method for transferring pre-trained self-supervised (SSL) speech representations to multiple languages. There is an abundance of unannotated speech, so creating self-supervised representations from raw audio and fine-tuning on small annotated datasets is a promising direction to build speech recognition systems. SSL models generally perform SSL on raw audio in a pre-training phase and then fine-tune on a small fraction of annotated data. Such models have produced state of the art results for ASR. However, these models are very expensive to pre-train. We use an existing wav2vec 2.0 model and tackle the problem of learning new language representations while utilizing existing model knowledge. Crucially we do so without catastrophic forgetting of the existing language representation. We use adapter modules to speed up pre-training a new language task. Our model can decrease pre-training times by 32% when learning a new language task, and learn this new audio-language representation without forgetting previous language representation. We evaluate by applying these language representations to automatic speech recognition.

Dario Fuoli, Zhiwu Huang, Shuhang Gu et al.

This paper reviews the video extreme super-resolution challenge associated with the AIM 2020 workshop at ECCV 2020. Common scaling factors for learned video super-resolution (VSR) do not go beyond factor 4. Missing information can be restored well in this region, especially in HR videos, where the high-frequency content mostly consists of texture details. The task in this challenge is to upscale videos with an extreme factor of 16, which results in more serious degradations that also affect the structural integrity of the videos. A single pixel in the low-resolution (LR) domain corresponds to 256 pixels in the high-resolution (HR) domain. Due to this massive information loss, it is hard to accurately restore the missing information. Track 1 is set up to gauge the state-of-the-art for such a demanding task, where fidelity to the ground truth is measured by PSNR and SSIM. Perceptually higher quality can be achieved in trade-off for fidelity by generating plausible high-frequency content. Track 2 therefore aims at generating visually pleasing results, which are ranked according to human perception, evaluated by a user study. In contrast to single image super-resolution (SISR), VSR can benefit from additional information in the temporal domain. However, this also imposes an additional requirement, as the generated frames need to be consistent along time.