A Reliability Assessment of LALM Audio Judges for Full-Duplex Voice Agents
Quick Answer
The study evaluates the reliability of Gemini models (2.5 Flash, 3.5 Flash, 3.1 Pro) as audio judges for full-duplex conversations, showing that Gemini 2.5 Flash aligns closely with human raters across multiple dimensions.
Quick Take
The study evaluates the reliability of Gemini models (2.5 Flash, 3.5 Flash, 3.1 Pro) as audio judges for full-duplex conversations, showing that Gemini 2.5 Flash aligns closely with human raters across multiple dimensions. The findings suggest that LALM can serve as a cost-effective substitute for human ratings, with significant implications for deployment in audio evaluation tasks.
Key Points
- Gemini 2.5 Flash shows high reliability, aligning with human raters in 6 of 8 dimensions.
- LALM's performance is consistent across tests, with minimal deviation from human scores.
- 3.5 Flash improves agreement across all dimensions, while 3.1 Pro rates lower despite similar rank correlation.
- Human rating costs are estimated to be two orders of magnitude higher than LALM workloads.
- Deployment requires careful consideration in four identified areas.
Paper Resources
📖 Reader Mode
~2 min readAbstract:We report the empirical reliability of Gemini models as audio judges that score full-duplex agent conversations directly from the raw stereo waveform, tested across three models in the Gemini family: 2.5 Flash, 3.5 Flash, and 3.1 Pro. Our primary evidence base uses Gemini 2.5 Flash as the ground-truth model, validated against three calibrated human raters on 209 stereo sessions, scored on 8 production dimensions: 152 full-duplex conversations across 13 accent-and-condition strata, together with 57 adversarial defect-injected clips. The evidence for Gemini 2.5 Flash is consistent across three tests. (i) On 5 of 8 dimensions the LALM-human Spearman rho departs from the pairwise human-human rho by at most 0.07, and on 7 of 8 dimensions the two quantities 95 percent bootstrap confidence intervals overlap. (ii) The LALM agrees with the three-rater human mean within 1 point on 60 to 92 percent of sessions on 6 of 8 dimensions. (iii) On 45 of 48 (defect, dimension) cells the LALM is as sensitive as humans or better under Newcombe-Wilson 95 percent confidence intervals, though most of these are underpowered nulls rather than demonstrated parity. Rank-ordering ability transfers across the Gemini family: 3.5 Flash improves simple agreement to 8 of 8 dimensions, while 3.1 Pro rates several dimensions markedly lower than humans despite comparable rank correlation. A model swap should be re-validated on calibration specifically, not assumed from rank-correlation alone. We identify four areas where deployment requires care, and we estimate that human rating alone for our current evaluation cadence costs roughly two orders of magnitude more than the equivalent LALM workload. The data presented here provides a defensible empirical basis for deploying the LALM as a substitute or fourth rater on the dimensions where the evidence supports it.
| Comments: | 28 pages total (12 main body, 1 reference, 15 appendix). In main body: 2 diagrams, 3 table, 2 charts |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS) |
| Cite as: | arXiv:2607.07985 [cs.CL] |
| (or arXiv:2607.07985v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.07985 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Armaan Sayyad [view email]
[v1]
Wed, 8 Jul 2026 23:24:55 UTC (389 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.