BFCL Audio: An Audio Function Calling Evaluation for Large Language Models

Huanzhi Mao, Aditya Ghai, Imra Dawoodani, Tony A Ginart, Shishir G Patil, John Emmons, Joseph E. Gonzalez
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:86410-86428, 2026.

Abstract

Audio agents are increasingly deployed to execute tools from spoken requests, yet audio tool use poses challenges beyond text-only function calling: perception errors (e.g., homophones, noise, disfluencies) can corrupt entities and arguments, and natural interactions often require clarification that changes the tool-calling protocol. We introduce BFCL Audio, a large-scale benchmark for audio function calling with 6.2K expert-verified tasks across two suites that mirror common deployments: BFCL Text Audio (pipelined $\text{ASR}\rightarrow\text{LLM}\rightarrow\text{tools}$ via transcripts) and BFCL True Audio (end-to-end $\text{audio-in}\rightarrow\text{tool calls}$). BFCL Audio includes controlled speech and acoustic perturbations (accent and speaking-rate variation, content disfluencies, and background noise) generated through a controllable audio synthesis/augmentation pipeline. We provide automatic grading for both function names and argument values using AST-based matching for single-turn calls and response/state-based metrics for multi-turn interactions, enabling scalable evaluation without LLM judges. Across a broad set of models, we propose a failure-mode taxonomy and analyze which speech and noise factors most strongly impact tool-calling accuracy. We release the benchmark, evaluation harness, and audio pipeline to support research on reliable speech-based agents.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-mao26b, title = {{BFCL} Audio: An Audio Function Calling Evaluation for Large Language Models}, author = {Mao, Huanzhi and Ghai, Aditya and Dawoodani, Imra and Ginart, Tony A and Patil, Shishir G and Emmons, John and Gonzalez, Joseph E.}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {86410--86428}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/mao26b/mao26b.pdf}, url = {https://proceedings.mlr.press/v306/mao26b.html}, abstract = {Audio agents are increasingly deployed to execute tools from spoken requests, yet audio tool use poses challenges beyond text-only function calling: perception errors (e.g., homophones, noise, disfluencies) can corrupt entities and arguments, and natural interactions often require clarification that changes the tool-calling protocol. We introduce BFCL Audio, a large-scale benchmark for audio function calling with 6.2K expert-verified tasks across two suites that mirror common deployments: BFCL Text Audio (pipelined $\text{ASR}\rightarrow\text{LLM}\rightarrow\text{tools}$ via transcripts) and BFCL True Audio (end-to-end $\text{audio-in}\rightarrow\text{tool calls}$). BFCL Audio includes controlled speech and acoustic perturbations (accent and speaking-rate variation, content disfluencies, and background noise) generated through a controllable audio synthesis/augmentation pipeline. We provide automatic grading for both function names and argument values using AST-based matching for single-turn calls and response/state-based metrics for multi-turn interactions, enabling scalable evaluation without LLM judges. Across a broad set of models, we propose a failure-mode taxonomy and analyze which speech and noise factors most strongly impact tool-calling accuracy. We release the benchmark, evaluation harness, and audio pipeline to support research on reliable speech-based agents.} }
Endnote
%0 Conference Paper %T BFCL Audio: An Audio Function Calling Evaluation for Large Language Models %A Huanzhi Mao %A Aditya Ghai %A Imra Dawoodani %A Tony A Ginart %A Shishir G Patil %A John Emmons %A Joseph E. Gonzalez %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-mao26b %I PMLR %P 86410--86428 %U https://proceedings.mlr.press/v306/mao26b.html %V 306 %X Audio agents are increasingly deployed to execute tools from spoken requests, yet audio tool use poses challenges beyond text-only function calling: perception errors (e.g., homophones, noise, disfluencies) can corrupt entities and arguments, and natural interactions often require clarification that changes the tool-calling protocol. We introduce BFCL Audio, a large-scale benchmark for audio function calling with 6.2K expert-verified tasks across two suites that mirror common deployments: BFCL Text Audio (pipelined $\text{ASR}\rightarrow\text{LLM}\rightarrow\text{tools}$ via transcripts) and BFCL True Audio (end-to-end $\text{audio-in}\rightarrow\text{tool calls}$). BFCL Audio includes controlled speech and acoustic perturbations (accent and speaking-rate variation, content disfluencies, and background noise) generated through a controllable audio synthesis/augmentation pipeline. We provide automatic grading for both function names and argument values using AST-based matching for single-turn calls and response/state-based metrics for multi-turn interactions, enabling scalable evaluation without LLM judges. Across a broad set of models, we propose a failure-mode taxonomy and analyze which speech and noise factors most strongly impact tool-calling accuracy. We release the benchmark, evaluation harness, and audio pipeline to support research on reliable speech-based agents.
APA
Mao, H., Ghai, A., Dawoodani, I., Ginart, T.A., Patil, S.G., Emmons, J. & Gonzalez, J.E.. (2026). BFCL Audio: An Audio Function Calling Evaluation for Large Language Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:86410-86428 Available from https://proceedings.mlr.press/v306/mao26b.html.

Related Material