OpenSound/CapSpeech

CapSpeech 是一个用于风格字幕文本到语音（CapTTS）任务的基准数据集，包括带有声音效果的 CapTTS 合成（CapTTS-SE）、口音字幕文本到语音（AccCapTTS）、情感字幕文本到语音（EmoCapTTS）以及聊天代理的文本到语音合成（AgentTTS）。CapSpeech 包含超过 1000 万个机器注释的音频字幕对和近 36 万个人工注释的音频字幕对。该数据集针对 CapTTS-SE 和 AgentTTS 任务设计了 3 个新的语音数据集，以增强基准数据集对现实场景的覆盖范围。

CapSpeech is a benchmark dataset designed for style-captioned text-to-speech (CapTTS) tasks, including CapTTS synthesis with sound effects (CapTTS-SE), accent-captioned TTS (AccCapTTS), emotion-captioned TTS (EmoCapTTS), and text-to-speech synthesis for chat agents (AgentTTS). The dataset comprises over 10 million machine-annotated audio-caption pairs and nearly 0.36 million human-annotated audio-caption pairs. It is structured with various splits for pretraining (PT) and supervised fine-tuning (SFT) stages, and is designed to enhance the benchmarks coverage of real-world scenarios by providing detailed annotations and support for different speech styles and effects.