High-quality International Phonetic Alphabet (IPA) transcriptions are a foundational resource for speech and language technologies, yet existing tools for many low-resource languages remain limited in accuracy and scope. In this work, we present MekongPhon, a large-scale, high-quality parallel IPA corpus for Lao and Khmer. The corpus contains 1.3 million Khmer and 367 thousand Lao orthographic–IPA pairs, meticulously aligned and verified. When used to train Transformer-based sequence-to-sequence models, MekongPhon enables exceptionally accurate IPA generation, achieving under 2% Character Error Rate (CER) on held-out test sets. We further introduce linguistically informed Lao and Khmer transliteration tools that offer high-speed IPA conversion, outperforming Epitran by 6-71 CER points despite trading some accuracy for efficiency. All data, code, and pretrained models are publicly released to support future research and development in low-resource language technologies.

Multilingual Neural Machine Translation (MNMT) models enhance translation quality for low-resource languages by exploiting cross-lingual similarities during training—a process known as knowledge transfer. This transfer is particularly effective between languages that share lexical or structural features, often enabled by a common orthography. However, languages with strong phonetic and lexical similarities but distinct writing systems experience limited benefits, as the absence of a shared orthography hinders knowledge transfer. To address this limitation, we propose an approach based on phonetic information that enhances token-level alignment across scripts by leveraging transliterations. We systematically evaluate several phonetic transcription techniques and strategies for incorporating phonetic information into NMT models. Our results show that using a shared encoder to process orthographic and phonetic inputs separately consistently yields the best performance for Khmer, Thai, and Lao in both directions with English, and that our custom Cognate-Aware Transliteration (CAT) method consistently improves translation quality over the baseline.