pere/nb-asr-numerics-balanced

Balanced Synthetic Norwegian Bokmål Numerics Dataset

This dataset provides a class-balanced synthetic corpus of Norwegian Bokmål sentences containing numeric expressions. It draws 10,000 examples for each of the 59 numeric categories (totaling 590,000 rows).

Source & Synthesis Architecture

• —Templates source: pere/nb-asr-numerics-categorized.
• —Methodology:
• —Filtered the original dataset for kept rows containing annotated entities.
• —For each target category, sampled 10,000 template sentences with replacement.
• —Substituted the <NUM> placeholders in each sentence template with generated random values matching the specified regex patterns for each category.
• —All entity boundaries, start/end offsets, and text values were deterministic-recalculated and aligned to match the newly generated text.
• —This approach yields a highly balanced corpus with maximized variance for downstream TTS pronunciation variant training.

Schema Details

• —id: Unique synthetic identifier.
• —text: Newly generated sentence text with synthetic numeric values.
• —masked_text: Original template text with <NUM> placeholders.
• —decision: Always "keep".
• —entities: List of aligned semantic entities with their start/end offsets, updated text, and categories.
• —constraints: Relational constraints from the source template.
• —primary_category: The category under which this row was generated.
• —source_template_id: The ID of the original template row.