aiqbench analysis
This arXiv paper proposes selecting small, diverse training subsets for fine-tuning Whisper speech recognition models. It uses Whisper's pretrained encoder to embed audio and a greedy k-center algorithm to pick representative samples, and reports gains over random sampling for Hebrew and Danish, mostly at small data budgets. For practitioners adapting ASR to low-resource languages, the practical takeaway is that data selection can reduce fine-tuning cost, though the results cover only two languages and the lexical-diversity explanation is correlational.
Whispering Wisely: Smart Subset Selection for ASR Fine-Tuning
arXiv:2610.32997v1 Announce Type: cross Abstract: Automatic Speech Recognition (ASR) models such as OpenAIs Whisper have substantially improved speech recognition quality but remain expensive to fine-tune, especially when large amounts of transcribed audio are available. This makes it challenging for many practitioners to adapt these models to new languages or domains. In this work, we introduce an embedding-based subset selection approach for selecting smaller training sets for ASR fine-tuning. Using Whispers pretrained encoder, we map each audio sample into a fixed-dimensional embedding space and apply a greedy k-center algorithm to select diverse and representative subsets of the training data. We evaluate this approach on both Hebrew and Danish, showing that k-center selection often improves upon random sampling, particularly at smaller subset sizes. An analysis of lexical diversity reveals that k-center subsets capture a broader range of linguistic content than random subsets of the same size, which may help explain their observed effectiveness. Overall, our results demonstrate that embedding-based subset selection provides a practical strategy for ASR subset selection in the two evaluated languages, with the clearest recognition benefits observed at smaller sample budgets.
Abstract:Automatic Speech Recognition (ASR) models such as OpenAIs Whisper have substantially improved speech recognition quality but remain expensive to fine-tune, especially when large amounts of transcribed audio are available. This makes it challenging for many practitioners to adapt these models to new languages or domains. In this work, we introduce an embedding-based subset selection approach for selecting smaller training sets for ASR fine-tuning. Using Whispers pretrained encoder, we map each audio sample into a fixed-dimensional embedding space and apply a greedy k-center algorithm to select diverse and representative subsets of the training data. We evaluate this approach on both Hebrew and Danish, showing that k-center selection often improves upon random sampling, particularly at smaller subset sizes. An analysis of lexical diversity reveals that k-center subsets capture a broader range of linguistic content than random subsets of the same size, which may help explain their observed effectiveness. Overall, our results demonstrate that embedding-based subset selection provides a practical strategy for ASR subset selection in the two evaluated languages, with the clearest recognition benefits observed at smaller sample budgets.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.32997 [cs.CL]
(or arXiv:2610.32997v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.32997 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 273 words