Same Words, Different Judgments: Modality Effects on Preference Alignment

• URL: http://arxiv.org/abs/2602.22710v1
• Date: Thu, 26 Feb 2026 07:34:15 GMT
• Title: Same Words, Different Judgments: Modality Effects on Preference Alignment
• Abstract summary: We compare text and audio evaluations of identical semantic content across 100 prompts.<n>Audio preferences prove as reliable as text, with inter-rater agreement reaching good levels.<n>Audio raters exhibit narrower decision thresholds, reduced length bias, and more user-oriented evaluation criteria.
• Score: 8.352948546053776
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Preference-based reinforcement learning (PbRL) is the dominant framework for aligning AI systems to human preferences, but its application to speech remains underexplored. We present a controlled cross-modal study of human and synthetic preference annotations, comparing text and audio evaluations of identical semantic content across 100 prompts. Audio preferences prove as reliable as text, with inter-rater agreement reaching good levels (ICC(2,k) $\approx$ .80) at $\sim$9 raters -- the first ICC-based reliability characterization in the preference annotation literature for either modality. However, modality reshapes how people judge: audio raters exhibit narrower decision thresholds, reduced length bias, and more user-oriented evaluation criteria, with near-chance cross-modality agreement. Synthetic ratings further align with human judgments and predict inter-rater agreement, supporting their use both for triaging ambiguous pairs and as full replacements for human annotations.

Related papers

• ECHO: A Matched-Contrast Benchmark for Context-Sensitive Turn-Taking in Full-Duplex Dialogue [19.857222937196745]
  We develop diagnostic benchmark for Chinese full spoken dialogue turn-taking pairs pairs examples.<n>We introduce pair accuracy, which requires correct decisions for both members of a pair and assigns no credit to constant-action policies.<n>These findings demonstrate that interruption-only evaluation can distinguish practical turntaking reliability.
  arXiv  Detail & Related papers  (2026-09-15T15:58:48Z)
• VA-Judger: Reward Modeling from Human Preference Feedback for Joint Video-Audio Generation [79.34043102155964]
  We propose VA-Judger, a chain-of-thought omni-reward model for joint video-audio generation.<n>We show VA-Judger outperforms metric baselines in predicting human preferences on both in-domain and out-of-domain evaluations.
  arXiv  Detail & Related papers  (2026-08-19T06:53:14Z)
• CriterAlign: Criterion-Centric Rationale Alignment for Code Preference Judging [95.02210956333374]
  We propose a criterion-centric framework that adapts rubric-based judging to pairwise preference evaluation.<n>On BigCodeReward, CriterAlign improves a Qwen2.5-VL-32B monolithic judge from 60.4% to 66.3% accuracy.
  arXiv  Detail & Related papers  (2026-05-19T10:59:19Z)
• AQAScore: Evaluating Semantic Alignment in Text-to-Audio Generation via Audio Question Answering [97.52852990265136]
  We introduce AQAScore, a backbone-agnostic evaluation framework that leverages the reasoning capabilities of audio-aware large language models.<n>We evaluate AQAScore across multiple benchmarks, including human-rated relevance, pairwise comparison, and compositional reasoning tasks.
  arXiv  Detail & Related papers  (2026-01-21T07:35:36Z)
• ORCA: Open-ended Response Correctness Assessment for Audio Question Answering [41.72231074041232]
  We present ORCA, a framework that models the variability in human judgments using Beta distributions to predict both expected correctness and uncertainty.<n>We collected 11,721 annotations across 3,580 question-answer pairs from 15 LALMs on two audio QA benchmarks, achieving inter-annotator agreement of 0.82 (Krippendorff's alpha)
  arXiv  Detail & Related papers  (2025-11-28T14:41:48Z)
• Aligning Text-to-Music Evaluation with Human Preferences [63.08368388389259]
  We study the design space of reference-based divergence metrics for evaluating generative acoustic text-to-music (TTM) models.<n>We find that not only is the standard FAD setup inconsistent on both synthetic and human preference data, but that nearly all existing metrics fail to effectively capture desiderata.<n>We propose a new metric, the MAUVE Audio Divergence (MAD), computed on representations from a self-supervised audio embedding model.
  arXiv  Detail & Related papers  (2025-03-20T19:31:04Z)
• Mind the Gap! Static and Interactive Evaluations of Large Audio Models [55.87220295533817]
  Large Audio Models (LAMs) are designed to power voice-native experiences.<n>This study introduces an interactive approach to evaluate LAMs and collect 7,500 LAM interactions from 484 participants.
  arXiv  Detail & Related papers  (2025-02-21T20:29:02Z)
• Speechworthy Instruction-tuned Language Models [71.8586707840169]
  We show that both prompting and preference learning increase the speech-suitability of popular instruction-tuned LLMs. We share lexical, syntactical, and qualitative analyses to showcase how each method contributes to improving the speech-suitability of generated responses.
  arXiv  Detail & Related papers  (2024-09-23T02:34:42Z)
• CLAIR-A: Leveraging Large Language Models to Judge Audio Captions [73.51087998971418]
  evaluating machine-generated audio captions is a complex task that requires considering diverse factors.<n>We propose CLAIR-A, a simple and flexible method that leverages the zero-shot capabilities of large language models.<n>In our evaluations, CLAIR-A better predicts human judgements of quality compared to traditional metrics.
  arXiv  Detail & Related papers  (2024-09-19T17:59:52Z)
• ROPO: Robust Preference Optimization for Large Language Models [59.10763211091664]
  We propose an iterative alignment approach that integrates noise-tolerance and filtering of noisy samples without the aid of external models. Experiments on three widely-used datasets with Mistral-7B and Llama-2-7B demonstrate that ROPO significantly outperforms existing preference alignment methods.
  arXiv  Detail & Related papers  (2024-04-05T13:58:51Z)
• Learning Personalized Alignment for Evaluating Open-ended Text Generation [44.565686959174585]
  PerSE is an interpretable evaluation framework designed to assess alignment with specific human preferences. It is tuned to infer specific preferences from an in-context personal profile and evaluate the alignment between the generated content and personal preferences. Our 13B LLaMA-2-based PerSE shows a 15.8% increase in Kendall correlation and a 13.7% rise in accuracy with zero-shot reviewers.
  arXiv  Detail & Related papers  (2023-10-05T04:15:48Z)
• TRScore: A Novel GPT-based Readability Scorer for ASR Segmentation and Punctuation model evaluation and selection [1.4720080476520687]
  Punctuation and readability are key to readability in Automatic Speech Recognition. Human evaluation is expensive, time-consuming, and suffers from large inter-observer variability. We present TRScore, a novel readability measure using the GPT model to evaluate different segmentation and punctuation systems.
  arXiv  Detail & Related papers  (2022-10-27T01:11:32Z)
• Using multiple reference audios and style embedding constraints for speech synthesis [68.62945852651383]
  The proposed model can improve the speech naturalness and content quality with multiple reference audios. The model can also outperform the baseline model in ABX preference tests of style similarity.
  arXiv  Detail & Related papers  (2021-10-09T04:24:29Z)

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.