Summary of Enhancing Out-of-vocabulary Performance Of Indian Tts Systems For Practical Applications Through Low-effort Data Strategies, by Srija Anand et al.
Enhancing Out-of-Vocabulary Performance of Indian TTS Systems for Practical Applications through Low-Effort Data Strategies
by Srija Anand, Praveen Srinivasa Varadhan, Ashwin Sankar, Giri Raju, Mitesh M. Khapra
First submitted to arxiv on: 18 Jul 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 The paper proposes a solution to improve text-to-speech (TTS) systems for low-resource languages like Hindi and Tamil, which typically have limited training datasets. The authors highlight the issue of out-of-vocabulary (OOV) words in downstream applications, such as code-mixing with English. A benchmark is created to test the performance of state-of-the-art TTS systems on OOV words, revealing poor results. To address this limitation, a low-effort strategy is proposed to obtain more training data using volunteers instead of professional voice artists. The approach aims to improve OOV performance while maintaining voice quality and in-domain accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at a problem with text-to-speech systems for languages like Hindi and Tamil. These systems can’t understand words that aren’t in their training data, which is a big issue when we mix languages or use domain-specific vocabulary. The authors create a test to see how well these systems do on out-of-vocabulary words and find that they don’t do very well. To fix this, the researchers suggest getting more training data by having volunteers record words that aren’t in the original data. This approach aims to improve performance on unfamiliar words without affecting the overall quality of the system. |
