VoiceAssistant-Eval
收藏资源简介:
VoiceAssistant-Eval是一个全面的人工智能助手评估基准,旨在评估人工智能助手在听、说、看三个方面的能力。该数据集包含了10497个精心挑选的示例,涵盖了13个任务类别。这些任务包括自然声音、音乐和口语对话的听;多轮对话、角色扮演模仿和各种场景的说;以及高度异构的图像的看。VoiceAssistant-Eval旨在解决现有基准在个性化声音模仿、免提交互、日常生活中的各种音频上下文和视听整合评估等方面的不足。
VoiceAssistant-Eval is a comprehensive benchmark for AI assistant evaluation, designed to assess the capabilities of AI assistants across three core dimensions: listening, speaking, and visual perception. This dataset includes 10,497 carefully curated examples spanning 13 task categories. The covered tasks involve listening to natural sounds, music and spoken dialogues; speaking in multi-turn conversations, role-play imitation and various scenarios; and perceiving highly heterogeneous images. VoiceAssistant-Eval aims to address the limitations of existing benchmarks in aspects including personalized voice imitation, hands-free interaction, diverse audio contexts in daily life, and audio-visual integration evaluation.




