Summary of Pslm: Parallel Generation Of Text and Speech with Llms For Low-latency Spoken Dialogue Systems, by Kentaro Mitsui et al.
PSLM: Parallel Generation of Text and Speech with LLMs for Low-Latency Spoken Dialogue Systems
by Kentaro Mitsui, Koh Mitsuda, Toshiaki Wakatsuki, Yukiya Hono, Kei Sawada
First submitted to arxiv on: 18 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 presents a solution to address the challenges of multimodal language models in spoken dialogue systems. Current models face two major issues: response generation latency and the need for written responses before generating spoken ones. To overcome these limitations, the study extends the input and output sequences of the model to support parallel text and speech generation. The approach is tested on spoken question answering tasks, showing improved latency while maintaining response quality. Furthermore, the study demonstrates that generating speech in multiple sequences can further reduce latency. Demo samples are available for evaluation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper aims to make spoken dialogue systems better by fixing two big problems with current language models. Right now, these models take too long to respond because they need to generate a written response first, and then convert it to spoken words. This takes even longer when the speech sequences are much longer than text ones. To solve this, the researchers extended the model’s input and output sequences so that it can generate both text and speech at the same time. They tested this on questions and answers, showing that their approach is faster while still giving good responses. They also found that breaking down spoken responses into smaller chunks helps even more. |
Keywords
» Artificial intelligence » Question answering
