Newest papers

SyRA-CS: Synthetic-to-Real Adaptation for Code-Switched ASR with Meta-Learning

Fadel Thior, Ahmer Tabassum, Haania Siddiqui, Hanan Aldarmaki, Thamar Solorio · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31415 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

El reconocimiento automático de habla (ASR) con cambio de código (CS) sigue siendo difícil porque el habla real de CS es escasa y el habla sintética no se ajusta plenamente. Investigamos si el habla sintética de CS es más útil para aprender una inicialización amigable a la adaptación que para el ajuste continuo ordinario. Proponemos SyRA-CS, un marco de adaptación sintética-a-real específico para parejas basado en el meta-aprendizaje de los reptiles. Para cada par de idiomas y modelo pre-entrenado, formamos el meta-entrenamiento en el subconjunto sintético correspondiente de CS-FLEURS y luego nos adaptamos utilizando una hora de lenguaje objetivo real etiquetado. Evaluamos cuatro configuraciones arquitectónicas: Encoder-CTC, FastConformer-NN

Machine translation (Meta M2M100); the English original follows.

Abstract

Code-switched (CS) automatic speech recognition (ASR) remains difficult because real CS speech is scarce and synthetic speech does not fully match it. We investigate whether synthetic CS speech is more useful for learning an adaptation-friendly initialization than for ordinary continual fine-tuning. We propose SyRA-CS, a pair-specific synthetic-to-real adaptation framework based on Reptile meta-learning. For each language pair and pretrained model, we meta-train on the corresponding synthetic CS-FLEURS subset and then adapt using one hour of labeled real target speech. We evaluate four architectural settings: Encoder-CTC, FastConformer-RNN-T, Attention Encoder-Decoder (AED) and Speech-LLM. Experiments cover on Arabic-, French-, German-, and Hindi-English, against synthetic-only transfer, real-only fine-tuning, and continual fine-tuning. SyRA-CS improves over continual fine-tuning on WER in all 16 model-pair comparisons and on CER in 15 of 16, with relative macro WER reductions of 17.7% for XLS-R and 5.3-5.8% for the other three models. The gains hold across the tested models but vary in size across architectures, language pairs and metrics.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Speech recognition, Initialization, Adaptation (eye), Language model, Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange