Hebrew grapheme-to-phoneme conversion with ReNikud

ReNikud Plus is an ONNX-based Python package that converts unvocalized Hebrew to IPA and niqqud for TTS and research workflows. The model weights are approximately 310 MB and download automatically from Hugging Face on first use; the package was uploaded to PyPI on Jul 17, 2026.

• ReNikud Plus provides two outputs: IPA via phonemize and pointed Hebrew (niqqud) via vocalize, with phonemize preserving stress and vocalize omitting some stress marks.
• Model weights (~310 MB) download from Hugging Face on first run, or users can pass a local ONNX file to the API.
• ReNikud Plus was published on PyPI on Jul 17, 2026, and the accompanying research citation is Melichov et al., ReNikud (arXiv:2606.20179, 2026).
• Evidano can ingest ReNikud phonemized IPA and niqqud fields for thematic coding, frequency counts, co-occurrence analysis, and reproducible visual reporting.

ReNikud Plus is a Python package that runs ONNX inference to convert unvocalized Hebrew text into IPA or niqqud. See the original release page at PyPI.

• Primary use: phonemization for TTS, speech tech, and spoken-language research.
• Model weights (~310 MB) download automatically from Hugging Face on first run.
• Citation: Melichov et al., ReNikud (arXiv:2606.20179, 2026).

ReNikud Plus exposes a small API (G2P class) that runs an ONNX model to map Hebrew graphemes to phonemes. Example usage:

• g2p = G2P(); g2p.phonemize("שלום עולם") → ʃlˈom ʔolˈam
• g2p.vocalize(...) returns pointed Hebrew (niqqud) suitable for TTS engines that accept diacritics.
• If you already host weights, pass a local path: G2P('model.onnx').

Practical caveats: niqqud output is phonetically faithful but not 'publication-grade' and omits some diacritics (for example, shva in clusters). The package supports optional gender-conditioned decoding, which can be useful when modeling speaker-typical vowel realization.

Follow these steps to operationalize ReNikud outputs into an AI-enabled qualitative pipeline:

• Install and run ReNikud (pip install renikud-plus); cache ONNX weights locally (~310 MB).
• Phonemize your unvocalized corpus and save IPA and/or niqqud as separate fields.
• Import transcripts and phoneme fields into Evidano (text or CSV).
• Auto-code with Evidano to generate initial themes, then add manual codes where phonetics matter.
• Run frequency, co-occurrence, and cross-segment analyses, for example by speaker gender, region, or condition.
• Visualize results with word clouds and co-occurrence networks, and extract example quotes linked to IPA contexts.
• If needed, collect targeted follow-ups with Evidano AI avatars to probe phonetic or lexical patterns.

Automated phonemization is a research tool, not a clinical or definitive linguistic judgment. Validate model outputs on a representative sample of your corpus, especially when working with dialectal Hebrew or sociophonetic studies. For citation and model details see the package page on PyPI and the accompanying paper on arXiv.

ReNikud Plus (PyPI release Jul 17, 2026) lowers the barrier to producing IPA and niqqud from unvocalized Hebrew. For teams that analyze Hebrew speech or combine phonetics with thematic coding, pairing ReNikud outputs with Evidano accelerates synthesis, cross-segment comparisons, and reproducible visual reporting.

Ready to try it on your corpus? Import your ReNikud phoneme fields into Evidano and run thematic and frequency analyses in minutes: Try Evidano for free.