From the documentation, this app includes a high-level text normalization pipeline to improve pronunciation before inference.
The documentation describes language-specific normalization in Kotlin (TextNormalizer.kt and NumberUtils.kt) and sentence splitting/chunking in Rust (rust/src/lang/*). This is intended to prevent numbers, currencies, ranges, abbreviations, and other symbols from being pronounced incorrectly.
After further testing, I found that my previous assumption was incorrect.
Text Normalization is not disabled for all V2/V3 languages. It works correctly in other supported languages—the issue appears to affect Korean only.
When Korean is selected, the Text Normalization option becomes disabled (greyed out), so users cannot enable it.
Is this intentional, or is it a bug? Ideally, Korean should support the same text normalization pipeline as the other supported languages.
From the documentation, this app includes a high-level text normalization pipeline to improve pronunciation before inference.
The documentation describes language-specific normalization in Kotlin (TextNormalizer.kt and NumberUtils.kt) and sentence splitting/chunking in Rust (rust/src/lang/*). This is intended to prevent numbers, currencies, ranges, abbreviations, and other symbols from being pronounced incorrectly.
After further testing, I found that my previous assumption was incorrect.
Text Normalization is not disabled for all V2/V3 languages. It works correctly in other supported languages—the issue appears to affect Korean only.
When Korean is selected, the Text Normalization option becomes disabled (greyed out), so users cannot enable it.
Is this intentional, or is it a bug? Ideally, Korean should support the same text normalization pipeline as the other supported languages.