Generative Adversarial Networks (GAN) offer a promising approach for Neural
Machine Translation (NMT). However, feeding multiple morphologically languages
into a single model during training reduces the NMT's performance. In GAN,
similar to bilingual models, multilingual NMT only considers one reference
translation for each sentence during model training. This single reference
translation limits the GAN model from learning sufficient information about the
source sentence representation. Thus, in this article, we propose Denoising
Adversarial Auto-encoder-based Sentence Interpolation (DAASI) approach to
perform sentence interpolation by learning the intermediate latent
representation of the source and target sentences of multilingual language
pairs. Apart from latent representation, we also use the Wasserstein-GAN
approach for the multilingual NMT model by incorporating the model generated
sentences of multiple languages for reward computation. This computed reward
optimizes the performance of the GAN-based multilingual model in an effective
manner. We demonstrate the experiments on low-resource language pairs and find
that our approach outperforms the existing state-of-the-art approaches for
multilingual NMT with a performance gain of up to 4 BLEU points. Moreover, we
use our trained model on zero-shot language pairs under an unsupervised
scenario and show the robustness of the proposed approach.

本文提出了一个 DAASI 方法用于多语言语言对的句子插值， 该方法可通过学习源语言和目标语言的中间潜在表示来优化基于 GAN 的多语言翻译模型。实验证明，在低资源语言对上，该方法的表现超过了现有的最先进方法，性能提高了最多 4 BLEU 分数。

通过敌对训练利用多语言能力在低资源神经机器翻译中的应用

Exploiting Multilingualism in Low-resource Neural Machine Translation  via Adversarial Learning

We present a deep neural network based singing voice synthesizer, inspired by
the Deep Convolutions Generative Adversarial Networks (DCGAN) architecture and
optimized using the Wasserstein-GAN algorithm. We use vocoder parameters for
acoustic modelling, to separate the influence of pitch and timbre. This
facilitates the modelling of the large variability of pitch in the singing
voice. Our network takes a block of consecutive frame-wise linguistic and
fundamental frequency features, along with global singer identity as input and
outputs vocoder features, corresponding to the block of features. This
block-wise approach, along with the training methodology allows us to model
temporal dependencies within the features of the input block. For inference,
sequential blocks are concatenated using an overlap-add procedure. We show that
the performance of our model is competitive with regards to the
state-of-the-art and the original sample using objective metrics and a
subjective listening test. We also present examples of the synthesis on a
supplementary website and the source code via GitHub.

本研究提出了一种基于深度神经网络的歌声合成器，使用 Wasserstein-GAN 算法进行优化，通过声学建模的声码器参数，实现对歌唱声音中音高和音色的分离，采用分块方法建模输入块内的时间依赖关系，并使用重叠添加程序将连续块拼接在一起，该模型的绩效达到了与最先进技术相竞争的水平。