Name: Hebrew Nlp Voice Analysis
Author: Pickacho

Hebrew NLP & Voice Rules

Text Processing (NLP)

Morphological Complexity: Hebrew is highly agglutinative (prefixes like 'ב', 'ה', 'ו', 'כ' attach directly to nouns). Use specialized tokenizers and lemmatizers (e.g., hebrew-nlp, spaCy-he, Trankit, or Dicta APIs) before running core logic.
RTL Directionality: Ensure all text normalization handles Bidirectional (BiDi) characters correctly, removing Zero-Width Non-Joiners (ZWNJ) or layout markers if scraping.
LLM Prompting: When querying LLMs, specifically direct them to generate output in "Modern Native Hebrew without archaic structures, matching spoken Israeli phrasing."

Transcription Tools: Prefer OpenAI's Whisper model (or localized enterprise tools) which handles Hebrew speech contexts far better than legacy STT.
: Filter out background noise using VAD (Voice Activity Detection), as Hebrew speech cadence can be dense and highly intonated.

Morphological Complexity: Hebrew is highly agglutinative (prefixes like 'ב', 'ה', 'ו', 'כ' attach directly to nouns). Use specialized tokenizers and lemmatizers (e.g., hebrew-nlp, spaCy-he, Trankit, or Dicta APIs) before running core logic.
RTL Directionality: Ensure all text normalization handles Bidirectional (BiDi) characters correctly, removing Zero-Width Non-Joiners (ZWNJ) or layout markers if scraping.
LLM Prompting: When querying LLMs, specifically direct them to generate output in "Modern Native Hebrew without archaic structures, matching spoken Israeli phrasing."

Transcription Tools: Prefer OpenAI's Whisper model (or localized enterprise tools) which handles Hebrew speech contexts far better than legacy STT.
: Filter out background noise using VAD (Voice Activity Detection), as Hebrew speech cadence can be dense and highly intonated.