Topics of Interest

We welcome work that:

• Advances efficient adaptation of Speech LLMs for low-resource languages (e.g., PEFT, multilingual transfer, distillation, edge/streaming constraints)
• Proposes or validates task designs and metrics beyond WER for ASR, SLU, and speech generation, including calibration and abstention
• Characterises robustness under domain shift (accent/dialect, channel/codec/reverberation, spontaneous speech) and compares cascaded vs. end-to-end error propagation
• Creates or curates LR resources with rigorous documentation and balanced, slice-aware evaluation splits (incl. lexicon/G2P for unwritten languages)
• Normalises ethics-by-default reporting (data/model cards, synthetic-data disclosure, lightweight privacy/safety guardrails)

Submissions that couple methodological innovation with thorough empirical evidence, calibrated uncertainty, and openly released artefacts are particularly encouraged.

Diversity and Inclusion

Contributions to Academic Diversity

SPEAKABLE centers less-resourced languages and dialects. We encourage submissions that release resources/code where possible, provide honest data/model cards, and report slice-aware metrics (accent/dialect, channel, speaker). We explicitly welcome SLU/dialog metrics and speech generation evaluations, not only WER.