LIUM Submission for IWSLT 2026 Low-resource Speech Translation Track

2026-06-29 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

The LIUM submission for the IWSLT 2026 low-resource speech translation track details data augmentation methods for speech-to-text translation, specifically targeting scenarios with limited resources. It proposes two primary pipelines: pseudo-labeling and speech synthesis, aiming to generate parallel speech data without relying on human-annotated speech translation data. The work focuses on Central Kurdish–English language pairs, exploring the advantages and limitations of each augmentation technique. The most effective results were achieved using the pseudo-labeling pipeline, which yielded a BLEU score of 25.73 on the development set and 21.09 on the test set for Central Kurdish–English translation. This approach demonstrates a viable strategy for improving performance in challenging low-resource settings.

Key takeaway

For Machine Learning Engineers developing speech translation systems in low-resource environments, you should prioritize pseudo-labeling as a data augmentation strategy. This method demonstrated superior performance, achieving a BLEU score of 21.09 on the test set for Central Kurdish–English, without requiring extensive human-annotated data. Consider integrating pseudo-labeling into your pipeline to efficiently generate parallel speech data and improve model accuracy in similar challenging language pairs.

Key insights

Data augmentation via pseudo-labeling and speech synthesis improves low-resource speech translation.

Principles

• Pseudo-labeling outperforms speech synthesis for low-resource ST.
• Data augmentation can generate parallel speech data without human annotation.

Method

The proposed method involves two data augmentation pipelines: pseudo-labeling and speech synthesis, designed to create parallel speech data for low-resource speech-to-text translation, specifically for Central Kurdish–English.

In practice

• Apply pseudo-labeling for low-resource speech translation tasks.
• Explore synthetic data generation to reduce annotation reliance.

Topics

• Speech Translation
• Low-Resource NLP
• Data Augmentation
• Pseudo-labeling
• Speech Synthesis
• Central Kurdish-English

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation — Transfer learning and data augmentation effectively overcome scarce annotation resources for low-resource Chinese dialect discrimination.
• A Review of Pseudo-Labeling for Computer Vision — Pseudo-labeling extends beyond semi-supervised learning to enhance self-supervised and unsupervised methods.
• ReNikud: Audio-Supervised Hebrew Grapheme-to-Phoneme Conversion — ReNikud leverages weak audio supervision and a character-aligned pseudo-vocalization architecture for accurate Hebrew Grapheme-to-Phoneme conversion.
• Semi-Supervised Speech Confidence Detection using Pseudo-Labelling and Whisper Embeddings — A novel framework detects speaker confidence by fusing human-engineered features and Whisper embeddings, enhanced by pseudo-labelling for data scarcity.
• Multilingual Coreference Resolution via Cycle-Consistent Machine Translation — Coreference resolution in low-resource languages can be significantly improved by cycle-consistent machine translation data augmentation.
• Label-Aware Pseudo-Training Sample Generation for Text Classification — A new data augmentation method uses LLMs to subtly alter sentences, improving NLP performance with limited data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.