The NTTSU team’s submission leverages several large language models developed through a training procedure that includes continual pre-training and supervised fine-tuning. For paragraph-level translation, we generated synthetic paragraph-aligned data and utilized this data for training.In the task of translating Japanese to Chinese, we particularly focused on the speech domain translation. Specifically, we built Whisper models for Japanese automatic speech recognition (ASR). We used YODAS dataset for Whisper training. Since this data contained many noisy data pairs, we combined the Whisper outputs using ROVER for polishing the transcriptions. Furthermore, to enhance the robustness of the translation model against errors in the transcriptions, we performed data augmentation by forward translation from audio, using both ASR and base translation models.To select the best translation from multiple hypotheses of the models, we applied Minimum Bayes Risk decoding + reranking, incorporating scores such as COMET-QE, COMET, and cosine similarity by LaBSE.

A spoken language translation (ST) system consists of at least two modules: an automatic speech recognition (ASR) system and a machine translation (MT) system. In most cases, an MT is only trained and optimized using error-free text data. If the ASR makes errors, the translation accuracy will be greatly reduced. Existing studies have shown that training MT systems with ASR parameters or word lattices can improve the translation quality. However, such an extension requires a large change in standard MT systems, resulting in a complicated model that is hard to train. In this paper, a neural sequence-to-sequence ASR is used as feature processing that is trained to produce word posterior features given spoken utterances. The resulting probabilistic features are used to train a neural MT (NMT) with only a slight modification. Experimental results reveal that the proposed method improved up to 5.8 BLEU scores with synthesized speech or 4.3 BLEU scores with the natural speech in comparison with a conventional cascaded-based ST system that translates from the 1-best ASR candidates.

This paper is concerned with speech-to-speech translation that is sensitive to paralinguistic information. From the many different possible paralinguistic features to handle, in this paper we chose duration and power as a first step, proposing a method that can translate these features from input speech to the output speech in continuous space. This is done in a simple and language-independent fashion by training a regression model that maps source language duration and power information into the target language. We evaluate the proposed method on a digit translation task and show that paralinguistic information in input speech appears in output speech, and that this information can be used by target language speakers to detect emphasis.