Summary of On the Problem Of Text-to-speech Model Selection For Synthetic Data Generation in Automatic Speech Recognition, by Nick Rossenbach et al.
On the Problem of Text-To-Speech Model Selection for Synthetic Data Generation in Automatic Speech Recognition
by Nick Rossenbach, Ralf Schlüter, Sakriani Sakti
First submitted to arxiv on: 31 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 investigates the impact of various neural text-to-speech (TTS) systems on synthetic data creation for automatic speech recognition (ASR) training. By comparing five different TTS decoder architectures, the study finds that there is no clear correlation between the performance metrics used and the ASR results. Instead, it observes that autoregressive decoding outperforms non-autoregressive decoding in terms of NISQA MOS and intelligibility. The findings have implications for selecting suitable TTS systems for synthetic data generation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how different voice-generating AI models affect speech recognition technology. It tests five different AI models to see which one works best for creating fake audio data. Surprisingly, the quality of the generated audio doesn’t seem to affect how well a speech recognition system can understand it. However, autoregressive AI models did perform better than non-autoregressive ones in terms of sound quality and clarity. |
Keywords
» Artificial intelligence » Autoregressive » Decoder » Synthetic data
