Textual content Normalization (TN) is a key preprocessing step in Textual content-to-Speech (TTS) programs, changing written kinds into their canonical spoken equivalents. Conventional TN programs can exhibit excessive accuracy, however contain substantial engineering effort, are troublesome to scale, and pose challenges to language protection, notably in low-resource settings. We suggest PolyNorm, a prompt-based strategy 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 automated information curation and analysis, designed to facilitate scalable experimentation throughout numerous 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,
