Jianzong Wang

An Approach for Neural Machine Translation with Graph Attention Network

2020

The achievement of Neural Machine Translation (NMT) drew the attention of the professionals in recent years. The translation quality outperforms traditional methods such as Statistical Machine Translation. However, for the document-level machine translation tasks, the …

Dropout Regularization for Self-Supervised Learning of Transformer Encoder Speech Representation

2021

Predicting the altered acoustic frames is an effective way of self-supervised learning for speech representation.However, it is challenging to prevent the pretrained model from overfitting.In this paper, we proposed to introduce two dropout regularization methods …

Adaptive Activation Network For Low Resource Multilingual Speech Recognition

2022 · ArXiv.org

Low resource automatic speech recognition (ASR) is a useful but thorny task, since deep learning ASR models usually need huge amounts of training data. The existing models mostly established a bottleneck (BN) layer by pre-training …

Adapitch: Adaption Multi-Speaker Text-to-Speech Conditioned on Pitch Disentangling with Untranscribed Data

2022 · arXiv (Cornell University)

In this paper, we proposed Adapitch, a multi-speaker TTS method that makes adaptation of the supervised module with untranscribed data. We design two self supervised modules to train the text encoder and mel decoder separately …

Cocktail: Chunk-Adaptive Mixed-Precision Quantization for Long-Context LLM Inference

2025 · arXiv (Cornell University)

Recently, large language models (LLMs) have been able to handle longer and longer contexts. However, a context that is too long may cause intolerant inference latency and GPU memory usage. Existing methods propose mixed-precision quantization …