Summary of A Human-in-the-loop Approach to Improving Cross-text Prosody Transfer, by Himanshu Maurya and Atli Sigurgeirsson
A Human-in-the-Loop Approach to Improving Cross-Text Prosody Transfer
by Himanshu Maurya, Atli Sigurgeirsson
First submitted to arxiv on: 6 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 proposed Human-in-the-Loop (HitL) approach addresses the challenges of text-to-speech (TTS) prosody transfer models when faced with cross-text conditions. By conditioning on a reference utterance that differs from the target text, these models struggle to separate prosody from text, resulting in reduced perceived naturalness. To resolve this issue, HitL users adjust salient correlates of prosody to make the prosody more suitable for the target text while maintaining the overall reference prosodic effect. The adjusted renditions maintain the reference prosody and are rated as more appropriate for the target text 57.8% of the time. Our analysis suggests that limited user effort is sufficient for achieving these improvements, and that closeness in the latent reference space is not a reliable prosodic similarity metric for cross-text conditions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary TTS models can make different sounds for the same words by using a reference recording. But when the reference recording is different from the words being spoken, the model struggles to create a natural-sounding speech. To fix this problem, we ask humans to adjust certain parts of the sound to match the new text better. The human-adjusted sounds keep the overall tone and rhythm of the original recording but are rated as more suitable for the new text 57.8% of the time. It seems that a little bit of human effort can make a big difference in improving speech naturalness. |
