arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2411.07563cs.AI
收藏

通过大语言模型的上下文内知识检索改进字素到音素的转换

Improving Grapheme-to-Phoneme Conversion through In-Context Knowledge Retrieval with Large Language Models

Centre for Perceptual and Interactive Intelligence (CPII) Limited(感知与交互智能中心(CPII)有限公司);The Chinese University of Hong Kong(香港中文大学)

Dongrui Han, Mingyu Cui, Jiawen Kang, Xixin Wu, Xunying Liu, Helen Meng

AI总结:

针对字素到音素转换中的上下文歧义问题,提出融合大语言模型上下文内知识检索能力的转换系统,在Librig2p数据集上显著降低音素错误率,验证了该方法的有效性。

摘要:

字素到音素(G2P)转换是文本转语音(TTS)系统中的关键步骤,负责将字素映射为对应的语音表示。然而,该任务面临歧义问题:同一字素可根据上下文表示多个音素,这给G2P转换带来了挑战。受大语言模型(LLMs)在处理上下文感知场景中显著成功的启发,本文提出具备LLMs上下文内知识检索(ICKR)能力的上下文G2P转换系统,以提升消歧能力。研究在Librig2p数据集上充分验证了将ICKR融入G2P转换系统的有效性。具体而言,采用ICKR的最优上下文G2P转换系统性能优于基线,音素错误率(PER)的加权平均值绝对降低2.0%(相对降低28.9%)。在ICKR系统中使用GPT-4可在Librig2p数据集上实现3.5%的绝对提升(3.8%的相对提升)。

英文摘要:

Grapheme-to-phoneme (G2P) conversion is a crucial step in Text-to-Speech (TTS) systems, responsible for mapping grapheme to corresponding phonetic representations. However, it faces ambiguities problems where the same grapheme can represent multiple phonemes depending on contexts, posing a challenge for G2P conversion. Inspired by the remarkable success of Large Language Models (LLMs) in handling context-aware scenarios, contextual G2P conversion systems with LLMs' in-context knowledge retrieval (ICKR) capabilities are proposed to promote disambiguation capability. The efficacy of incorporating ICKR into G2P conversion systems is demonstrated thoroughly on the Librig2p dataset. In particular, the best contextual G2P conversion system using ICKR outperforms the baseline with weighted average phoneme error rate (PER) reductions of 2.0% absolute (28.9% relative). Using GPT-4 in the ICKR system can increase of 3.5% absolute (3.8% relative) on the Librig2p dataset.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑