Results of the summarizing process for the arXiv paper: 2304.03169v1

This paper proposes the use of an end-to-end (e2e) architecture for English-Hindi speech translation (ST), which has gained popularity over cascade systems due to its enhanced performance and ability to retain para and non-linguistic characteristics of speech. The authors use two imperfect machine translation (MT) services to translate Libri-trans en text into hi text, and propose a data augmentation strategy of noisy MT data to aid robust ST. The main contribution of this paper is the proposal of a selective data augmentation strategy that results in better ST performance, as measured by BLEU score, compared to brute force augmentation of MT data. The authors observed an absolute improvement of 1.59 BLEU score with their approach. The authors also show that more training data is not necessarily a good strategy for building an e2e ST model, and that selective augmentation of data helps improve performance in the task. They demonstrate that edit distance-based data selection improves system performance with a gain in BLEU score, even though the augmented data is less compared to using all available data. The generation of parallel data for low-resource en-Hindi is another contribution of this paper. The authors provide an example showing how their approach works by selecting training data from two sets based on edit distance between translations obtained via different MT services. They choose all the training data from one set and augment it with select data from the other set based on ED criteria, resulting in improved BLEU scores. Overall, this paper highlights the importance of selective augmentation strategies for improving e2e ST system performance and underscores the need for sufficiently large parallel corpora for Indian languages to promote further research in this direction.

• Proposal of an end-to-end architecture for English-Hindi speech translation (ST)
• Use of two imperfect machine translation (MT) services to translate Libri-trans en text into hi text
• Data augmentation strategy of noisy MT data to aid robust ST
• Selective data augmentation strategy proposed that results in better ST performance, as measured by BLEU score, compared to brute force augmentation of MT data
• Absolute improvement of 1.59 BLEU score observed with the proposed approach
• More training data is not necessarily a good strategy for building an e2e ST model, and that selective augmentation of data helps improve performance in the task
• Edit distance-based data selection improves system performance with a gain in BLEU score, even though the augmented data is less compared to using all available data
• Generation of parallel data for low-resource en-Hindi is another contribution of this paper
• Example provided showing how their approach works by selecting training data from two sets based on edit distance between translations obtained via different MT services
• Overall importance highlighted of selective augmentation strategies for improving e2e ST system performance and need for sufficiently large parallel corpora for Indian languages to promote further research in this direction.

End-to-End English-Hindi Speech Translation with Selective Data Augmentation

In recent years, end-to-end (e2e) architectures have gained popularity in speech translation (ST) due to their enhanced performance and ability to retain para and non-linguistic characteristics of speech. In this paper, the authors propose an e2e architecture for English-Hindi ST that uses two imperfect machine translation (MT) services to translate LibriSpeech en text into hi text. The main contribution of this paper is a selective data augmentation strategy that results in improved ST performance compared to brute force augmentation of MT data.

Data Augmentation Strategy

The authors use noisy MT data as a source for augmenting training data for the e2e model. To improve system performance, they propose a selective data augmentation strategy based on edit distance (ED). This approach involves selecting training data from two sets based on ED between translations obtained via different MT services. All the training data is chosen from one set and augmented with select data from the other set based on ED criteria.

The authors demonstrate that their approach works by showing an example where they choose all the training data from one set and augment it with select data from the other set based on ED criteria, resulting in improved BLEU scores. They observed an absolute improvement of 1.59 BLEU score with their approach compared to using all available MT data without selection or filtering. This highlights the importance of selective augmentation strategies for improving e2e ST system performance and underscores the need for sufficiently large parallel corpora for Indian languages to promote further research in this direction.

Generation of Parallel Data

Another contribution of this paper is its focus on generating parallel corpus for low resource en-Hindi language pairs which can be used as a source for building better models in future research efforts related to ST tasks involving Indian languages such as Hindi, Bengali etc.. The authors show how their proposed approach can be used effectively by selecting appropriate training samples from two sets based on edit distance between translations obtained via different MT services which helps improve system performance even though less amount of augmented dataset is used than when using all available dataset without selection or filtering .

Conclusion

Overall, this paper proposes an effective end-to-end architecture along with a novel selective augmentation strategy that improves English–Hindi speech translation performance significantly over existing methods while also providing insights into generating parallel corpus suitable for low resource language pairs such as en–hi pairings which could be useful in future research efforts related to Indian languages like Hindi or Bengali etc..