ReNikud: Audio-Supervised Hebrew Grapheme-to-Phoneme Conversion

2026-06-18 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

ReNikud, a novel method for Hebrew Grapheme-to-Phoneme (G2P) conversion, addresses significant challenges posed by the language's abjad writing system, which leaves vowels largely unwritten and creates ambiguity. Traditional G2P approaches struggle with scarce vocalization data, lack lexical stress specification, and reflect formal rules over spoken pronunciation. ReNikud overcomes these limitations through two key insights: weak audio supervision via a phoneme-based Automatic Speech Recognition (ASR) pseudo-labeling pipeline on thousands of hours of unlabeled Hebrew audio, yielding phonemic transcriptions reflecting natural spoken norms; and a pseudo-vocalization architecture that predicts IPA phonemes at each character position, enforcing character-level alignment. The method surpasses previous state-of-the-art on existing Hebrew G2P benchmarks and the new MILIM benchmark, with code and models slated for release.

Key takeaway

For NLP engineers developing Hebrew text-to-speech or speech technologies, ReNikud presents a significant advancement in Grapheme-to-Phoneme conversion. Its audio-supervised pseudo-labeling and character-aligned architecture address critical data and ambiguity challenges inherent in Hebrew's abjad system. You should explore integrating its methodology or utilizing the upcoming open-source models to enhance the naturalness and accuracy of your Hebrew speech applications.

Key insights

ReNikud leverages weak audio supervision and a character-aligned pseudo-vocalization architecture for accurate Hebrew Grapheme-to-Phoneme conversion.

Principles

• Weak audio supervision can mitigate data scarcity for G2P.
• ASR pseudo-labeling captures natural spoken pronunciation norms.
• Character-level alignment improves abjad G2P accuracy.

Method

ReNikud employs a phoneme-based ASR pseudo-labeling pipeline on thousands of hours of unlabeled Hebrew audio to generate phonemic transcriptions, then uses a pseudo-vocalization architecture to predict IPA phonemes at each character position.

In practice

• Improve naturalness of Hebrew text-to-speech systems.
• Advance development of Hebrew speech technologies.

Topics

• Hebrew G2P
• Text-to-Speech
• Automatic Speech Recognition
• Pseudo-labeling
• Phoneme Prediction
• Abjad Systems

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Phonikud: Overcoming Phonetic Underspecification for Hebrew Text-To-Speech — Fully specifying phonetic features like stress significantly improves Hebrew TTS accuracy and enables smaller models.
• From A to B to A: Palindromic Zero-Shot Voice Conversion with Non-Parallel Data — A voice conversion framework uses KNN over WavLM to create synthetic training pairs from non-parallel data for zero-shot, multilingual VC.
• Scaling Human and G2P Supervision for Robust Phonetic Transcription — G2P supervision for phonetic transcription has diminishing returns beyond 20-30 hours of human annotation.
• A Comparative Study of Pretrained Transformer Models for Quranic ASR: Speech Representations, Label Formats, and Dataset Composition — Fine-tuning pretrained Transformers with specific speech representations and label formats significantly improves Quranic ASR accuracy and efficiency.
• LoSATok: Low-dimensional Semantic-Acoustic Tokenizer for Cross-Domain Audio Understanding and Generation — LoSATok unifies audio understanding and generation via low-dimensional semantic-acoustic tokens, reducing Diffusion Transformer modeling complexity.
• FlowEdit: Associative Memory for Lifelong Pronunciation Adaptation in Flow-Matching TTS — FlowEdit enables lifelong TTS pronunciation adaptation via latent embedding edits stored in a Hopfield Network, reducing errors by 92.7%.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.