Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userContinual-wav2vec2: an Application of Continual Learning for Self-Supervised Automatic Speech Recognition
320
0
0.0
(
0
)
Ask ChatGPT about the research
No Arabic abstract
We present a method for continual learning of speech representations for multiple languages using self-supervised learning (SSL) and applying these for automatic speech recognition. There is an abundance of unannotated speech, so creating self-supervised representations from raw audio and finetuning on a small annotated datasets is a promising direction to build speech recognition systems. Wav2vec models perform SSL on raw audio in a pretraining phase and then finetune on a small fraction of annotated data. SSL models have produced state of the art results for ASR. However, these models are very expensive to pretrain with self-supervision. We tackle the problem of learning new language representations continually from audio without forgetting a previous language representation. We use ideas from continual learning to transfer knowledge from a previous task to speed up pretraining a new language task. Our continual-wav2vec2 model can decrease pretraining times by 32% when learning a new language task, and learn this new audio-language representation without forgetting previous language representation.
rate research
Read More
Despite the growing interest in unsupervised learning, extracting meaningful knowledge from unlabelled audio remains an open challenge. To take a step in this direction, we recently proposed a problem-agnostic speech encoder (PASE), that combines a convolutional encoder followed by multiple neural networks, called workers, tasked to solve self-supervised problems (i.e., ones that do not require manual annotations as ground truth). PASE was shown to capture relevant speech information, including speaker voice-print and phonemes. This paper proposes PASE+, an improved version of PASE for robust speech recognition in noisy and reverberant environments. To this end, we employ an online speech distortion module, that contaminates the input signals with a variety of random disturbances. We then propose a revised encoder that better learns short- and long-term speech dynamics with an efficient combination of recurrent and convolutional networks. Finally, we refine the set of workers used in self-supervision to encourage better cooperation. Results on TIMIT, DIRHA and CHiME-5 show that PASE+ significantly outperforms both the previous version of PASE as well as common acoustic features. Interestingly, PASE+ learns transferable representations suitable for highly mismatched acoustic conditions.
Stream fusion, also known as system combination, is a common technique in automatic speech recognition for traditional hybrid hidden Markov model approaches, yet mostly unexplored for modern deep neural network end-to-end model architectures. Here, we investigate various fusion techniques for the all-attention-based encoder-decoder architecture known as the transformer, striving to achieve optimal fusion by investigating different fusion levels in an example single-microphone setting with fusion of standard magnitude and phase features. We introduce a novel multi-encoder learning method that performs a weighted combination of two encoder-decoder multi-head attention outputs only during training. Employing then only the magnitude feature encoder in inference, we are able to show consistent improvement on Wall Street Journal (WSJ) with language model and on Librispeech, without increase in runtime or parameters. Combining two such multi-encoder trained models by a simple late fusion in inference, we achieve state-of-the-art performance for transformer-based models on WSJ with a significant WER reduction of 19% relative compared to the current benchmark approach.
For real-world speech recognition applications, noise robustness is still a challenge. In this work, we adopt the teacher-student (T/S) learning technique using a parallel clean and noisy corpus for improving automatic speech recognition (ASR) performance under multimedia noise. On top of that, we apply a logits selection method which only preserves the k highest values to prevent wrong emphasis of knowledge from the teacher and to reduce bandwidth needed for transferring data. We incorporate up to 8000 hours of untranscribed data for training and present our results on sequence trained models apart from cross entropy trained ones. The best sequence trained student model yields relative word error rate (WER) reductions of approximately 10.1%, 28.7% and 19.6% on our clean, simulated noisy and real test sets respectively comparing to a sequence trained teacher.
Acoustic models in real-time speech recognition systems typically stack multiple unidirectional LSTM layers to process the acoustic frames over time. Performance improvements over vanilla LSTM architectures have been reported by prepending a stack of frequency-LSTM (FLSTM) layers to the time LSTM. These FLSTM layers can learn a more robust input feature to the time LSTM layers by modeling time-frequency correlations in the acoustic input signals. A drawback of FLSTM based architectures however is that they operate at a predefined, and tuned, window size and stride, referred to as view in this paper. We present a simple and efficient modification by combining the outputs of multiple FLSTM stacks with different views, into a dimensionality reduced feature representation. The proposed multi-view FLSTM architecture allows to model a wider range of time-frequency correlations compared to an FLSTM model with single view. When trained on 50K hours of English far-field speech data with CTC loss followed by sMBR sequence training, we show that the multi-view FLSTM acoustic model provides relative Word Error Rate (WER) improvements of 3-7% for different speaker and acoustic environment scenarios over an optimized single FLSTM model, while retaining a similar computational footprint.
For artificial learning systems, continual learning over time from a stream of data is essential. The burgeoning studies on supervised continual learning have achieved great progress, while the study of catastrophic forgetting in unsupervised learning is still blank. Among unsupervised learning methods, self-supervise learning method shows tremendous potential on visual representation without any labeled data at scale. To improve the visual representation of self-supervised learning, larger and more varied data is needed. In the real world, unlabeled data is generated at all times. This circumstance provides a huge advantage for the learning of the self-supervised method. However, in the current paradigm, packing previous data and current data together and training it again is a waste of time and resources. Thus, a continual self-supervised learning method is badly needed. In this paper, we make the first attempt to implement the continual contrastive self-supervised learning by proposing a rehearsal method, which keeps a few exemplars from the previous data. Instead of directly combining saved exemplars with the current data set for training, we leverage self-supervised knowledge distillation to transfer contrastive information among previous data to the current network by mimicking similarity score distribution inferred by the old network over a set of saved exemplars. Moreover, we build an extra sample queue to assist the network to distinguish between previous and current data and prevent mutual interference while learning their own feature representation. Experimental results show that our method performs well on CIFAR100 and ImageNet-Sub. Compared with the baselines, which learning tasks without taking any technique, we improve the image classification top-1 accuracy by 1.60% on CIFAR100, 2.86% on ImageNet-Sub and 1.29% on ImageNet-Full under 10 incremental steps setting.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
