SER Evals Benchmark
收藏资源简介:
SER Evals Benchmark是由弗吉尼亚联邦大学和realiz.ai联合创建的一个大规模多语种语音情感识别数据集。该数据集包含17个不同语言和情感表达的数据集,旨在评估语音情感识别模型在不同领域和跨语言环境下的鲁棒性和泛化能力。数据集经过预处理和标签映射,确保了跨数据集的比较公平性。创建过程中,特别关注了较少使用的数据集,以减少过拟合并促进模型的鲁棒性。该数据集主要应用于人机交互领域,特别是在需要自然和同理心交互的场景中,以解决模型在多样语言和情感表达中的泛化问题。
The SER Evals Benchmark is a large-scale multilingual speech emotion recognition dataset co-developed by Virginia Commonwealth University and realiz.ai. This benchmark includes 17 datasets covering diverse languages and emotional expressions, aiming to evaluate the robustness and generalization capabilities of speech emotion recognition models across various domains and cross-lingual scenarios. All datasets within this benchmark have undergone preprocessing and label mapping to ensure fair cross-dataset comparisons. During its development, particular attention was paid to underutilized datasets to reduce overfitting and enhance model robustness. This benchmark is primarily applied in the field of human-computer interaction (HCI), especially in scenarios requiring natural and empathetic interactions, to address the generalization challenges of models when dealing with diverse languages and emotional expressions.

- 1SER Evals: In-domain and Out-of-domain Benchmarking for Speech Emotion Recognition弗吉尼亚联邦大学, 美国 · 2024年



