Overcoming Decoder Inconsistencies in Whisper for Dravidian and Low-Resource Languages

• URL: http://arxiv.org/abs/2606.09535v1
• Date: Mon, 08 Jun 2026 14:18:51 GMT
• Title: Overcoming Decoder Inconsistencies in Whisper for Dravidian and Low-Resource Languages
• Abstract summary: We show that Dravidian languages have longer words, higher vocabulary diversity, and lower repetition, resulting in sparse token distributions and frequent character-level substitution errors.<n>We introduce two decoder-level enhancements: Weighted-Attention, which adaptively balances attention sources, and Self-Conditioning, which reinjects intermediate predictions to improve token consistency.
• Score: 11.118322085806712
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multilingual ASR models such as Whisper perform well on high-resource languages but exhibit substantially higher Word Error Rates (WER) for Dravidian languages compared to Indo-Aryan ones. Through linguistic and dataset analysis, we show that Dravidian languages have longer words, higher vocabulary diversity, and lower repetition, resulting in sparse token distributions and frequent character-level substitution errors. Baseline fine-tuning further reveals decoder imbalance between self-attention (linguistic context) and cross-attention (acoustic cues). Although synthetic token-repetition experiments indicate potential gains, they are impractical. Motivated by these observations, we introduce two decoder-level enhancements: Weighted-Attention, which adaptively balances attention sources, and Self-Conditioning, which reinjects intermediate predictions to improve token consistency. Experiments demonstrate consistent WER reductions for low-resource and agglutinative languages.

Related papers

• Beyond Input Understanding: Diagnosing Multilingual Mathematical Reasoning with Directed Acyclic Trace Graphs [54.72295694649798]
  Large reasoning models (LRMs) achieve strong mathematical reasoning performance in English, but remain much less reliable in many low- and medium-resource languages.<n>We show that controlling the model's reasoning language can substantially reduce accuracy, suggesting that language also affects reasoning execution itself.<n>Experiments on the Qwen3 series across 12 languages show that non-English reasoning often suffers from reduced anchor coverage and weaker dependency fidelity.
  arXiv  Detail & Related papers  (2026-05-26T21:41:52Z)
• All Languages Matter: Understanding and Mitigating Language Bias in Multilingual RAG [87.74635133954975]
  We show that current Multilingual Retrieval-Augmented Generation (mRAG) systems suffer from a language bias during reranking.<n>We propose textittextbfLanguage-textbfAgnostic textbfUtility-driven textbfReranker textbfAlignment (LAURA), which aligns multilingual evidence ranking with downstream generative utility.
  arXiv  Detail & Related papers  (2026-04-22T05:33:06Z)
• Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages [39.985923582735936]
  We introduce the FRED Difficulty Metrics, which include the Fertility Ratio (F), Retrieval Proxy (R), Pre-training Exposure (E) and Corpus Diversity (D)<n>These metrics reveal that a significant portion of result variability is explained by train-test overlap and pre-training exposure rather than model capability.
  arXiv  Detail & Related papers  (2026-03-26T09:20:17Z)
• SITA: Learning Speaker-Invariant and Tone-Aware Speech Representations for Low-Resource Tonal Languages [11.655315357810371]
  SITA is a lightweight adaptation recipe that enforces Speaker-Invariance and Tone-Awareness for pretrained wav2vec-style encoders.<n>We evaluate primarily on Hmong, a highly tonal and severely under-the-shelf multilingual encoders fail to represent tone effectively.
  arXiv  Detail & Related papers  (2026-01-14T00:42:27Z)
• What Drives Cross-lingual Ranking? Retrieval Approaches with Multilingual Language Models [0.19116784879310025]
  Cross-lingual information retrieval is challenging due to disparities in resources, scripts, and weak cross-lingual semantic alignment in embedding models.<n>Existing pipelines often rely on translation and monolingual retrievals, which add computational overhead and noise, performance.<n>This work systematically evaluates four intervention types, namely document translation, multilingual dense retrieval with pretrained encoders, contrastive learning at word, phrase, and query-document levels, and cross-encoder re-ranking, across three benchmark datasets.
  arXiv  Detail & Related papers  (2025-11-24T17:17:40Z)
• Adapting Language Balance in Code-Switching Speech [60.296574524609575]
  Large foundational models still struggle against code-switching test cases.<n>We use differentiable surrogates to mitigate context bias during generation.<n>Experiments with Arabic and Chinese-English showed that the models are able to predict the switching places more correctly.
  arXiv  Detail & Related papers  (2025-10-21T15:23:55Z)
• Beyond WER: Probing Whisper's Sub-token Decoder Across Diverse Language Resource Levels [6.627057618324123]
  This paper introduces a fine-grained analysis of Whisper's multilingual decoder.<n>Our method traces the beam search path, capturing sub-token guesses and their associated probabilities.<n>Lower resource languages fare worse on these metrics, but also exhibit distinct clustering patterns in sub-token usage.
  arXiv  Detail & Related papers  (2025-09-29T21:20:05Z)
• Mechanistic Understanding and Mitigation of Language Confusion in English-Centric Large Language Models [56.61984030508691]
  We present the first mechanistic interpretability study of language confusion.<n>We show that confusion points (CPs) are central to this phenomenon.<n>We show that editing a small set of critical neurons, identified via comparative analysis with a multilingual-tuned counterpart, substantially mitigates confusion.
  arXiv  Detail & Related papers  (2025-05-22T11:29:17Z)
• Investigating and Scaling up Code-Switching for Multilingual Language Model Pre-Training [58.696660064190475]
  We find that the existence of code-switching, alternating between different languages within a context, is key to multilingual capabilities.<n>To better explore the power of code-switching for language alignment during pre-training, we investigate the strategy of synthetic code-switching.
  arXiv  Detail & Related papers  (2025-04-02T15:09:58Z)
• LAMASSU: Streaming Language-Agnostic Multilingual Speech Recognition and Translation Using Neural Transducers [71.76680102779765]
  Automatic speech recognition (ASR) and speech translation (ST) can both use neural transducers as the model structure. We propose LAMASSU, a streaming language-agnostic multilingual speech recognition and translation model using neural transducers.
  arXiv  Detail & Related papers  (2022-11-05T04:03:55Z)
• How Robust is Neural Machine Translation to Language Imbalance in Multilingual Tokenizer Training? [86.48323488619629]
  We analyze how translation performance changes as the data ratios among languages vary in the tokenizer training corpus. We find that while relatively better performance is often observed when languages are more equally sampled, the downstream performance is more robust to language imbalance than we usually expected.
  arXiv  Detail & Related papers  (2022-04-29T17:50:36Z)
• Detecting Fine-Grained Cross-Lingual Semantic Divergences without Supervision by Learning to Rank [28.910206570036593]
  This work improves the prediction and annotation of fine-grained semantic divergences. We introduce a training strategy for multilingual BERT models by learning to rank synthetic divergent examples of varying granularity. Learning to rank helps detect fine-grained sentence-level divergences more accurately than a strong sentence-level similarity model.
  arXiv  Detail & Related papers  (2020-10-07T21:26:20Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.