Textual content Normalization (TN) is a key preprocessing step in Textual content-to-Speech (TTS) techniques, changing written kinds into their canonical spoken equivalents. Conventional TN techniques can exhibit excessive accuracy, however contain substantial engineering effort, are tough to scale, and pose challenges to language protection, significantly in low-resource settings. We suggest PolyNorm, a prompt-based method to TN utilizing Massive Language Fashions (LLMs), aiming to cut back the reliance on manually crafted guidelines and allow broader linguistic applicability with minimal human intervention. Moreover, we current a language-agnostic pipeline for computerized knowledge curation and analysis, designed to facilitate scalable experimentation throughout various languages. Experiments throughout eight languages present constant reductions within the phrase error price (WER) in comparison with a production-grade-based system. To assist additional analysis,
