arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.23958eess.AScs.SD
收藏

Fréchet音频距离中任务诱导编码器偏差的实证分析

An Empirical Analysis of Task-Induced Encoder Bias in Fréchet Audio Distance

Dept. of Computer Science and Engineering, Sogang University, South Korea(计算机科学与工程系,首尔大学,韩国)

Wonwoo Jeong

AI总结:

通过分解评估指标为召回率、精度和对齐(语义与结构维度),分析六种编码器在FAD中的任务诱导偏差,发现重建、ASR和分类训练编码器各有优劣,需发展评估原生编码器。

摘要:

Fréchet音频距离(FAD)是评估文本到音频生成的事实标准,但其分数依赖于底层编码器的嵌入空间。编码器的训练任务决定了哪些声学特征被保留或丢弃,导致FAD继承系统性的任务诱导偏差。我们将评估分解为召回率、精度和对齐(分为语义和结构维度),并使用对数尺度归一化以实现公平的跨编码器比较。在两个数据集上对六种编码器进行的受控实验揭示了四轴权衡:基于重建的AudioMAE主导精度敏感性;ASR训练的Whisper在结构检测中占优,但对信号退化视而不见;分类训练的VGGish最大化语义检测,但惩罚合法的类内变异。由于没有单个编码器是通用评估器,未来的指标必须转向与人类感知内在一致的评估原生编码器。

英文摘要:

Fréchet Audio Distance (FAD) is the de facto standard for evaluating text-to-audio generation, yet its scores depend on the underlying encoder's embedding space. An encoder's training task dictates which acoustic features are preserved or discarded, causing FAD to inherit systematic task-induced biases. We decompose evaluation into Recall, Precision, and Alignment (split into semantic and structural dimensions), using log-scale normalization for fair cross-encoder comparison. Controlled experiments on six encoders across two datasets reveal a four-axis trade-off: reconstruction-based AudioMAE leads precision sensitivity; ASR-trained Whisper dominates structural detection but is blind to signal degradation; classification-trained VGGish maximizes semantic detection but penalizes legitimate intra-class variation. Since no single encoder is a universal evaluator, future metrics must shift toward evaluation-native encoders intrinsically aligned with human perception.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑