Summary of Costa: Code-switched Speech Translation Using Aligned Speech-text Interleaving, by Bhavani Shankar et al.
CoSTA: Code-Switched Speech Translation using Aligned Speech-Text Interleaving
by Bhavani Shankar, Preethi Jyothi, Pushpak Bhattacharyya
First submitted to arxiv on: 16 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG); Sound (cs.SD); Audio and Speech Processing (eess.AS)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper presents a novel approach to spoken translation of code-switched speech in Indian languages, such as Bengali, Hindi, Marathi, and Telugu, into English text. The authors propose an end-to-end model architecture called COSTA that leverages pre-trained automatic speech recognition (ASR) and machine translation (MT) modules. The fusion of speech and ASR text representations is achieved through an aligned interleaving scheme, which is then fed as input to a pre-trained MT module. The entire pipeline is trained end-to-end for spoken translation using synthetically created data. The authors also release a new evaluation benchmark for code-switched Bengali-English, Hindi-English, Marathi-English, and Telugu-English speech to English text. Experimental results show that COSTA significantly outperforms competitive cascaded and end-to-end multimodal baselines by up to 3.5 BLEU points. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to translate spoken language from different Indian languages (like Bengali or Hindi) into English. The problem is that there’s not enough data available for this kind of translation, making it hard to train good models. The authors created a new model called COSTA that combines two existing technologies: automatic speech recognition and machine translation. They trained their model using fake data and tested it against other similar models. The results show that COSTA is much better than the others at translating spoken language. |
Keywords
* Artificial intelligence * Bleu * Translation
