OpenSound/CapSpeech-PT

CapSpeech是一个用于风格标注文本到语音合成的基准数据集，包含超过1000万机器标注的音频-字幕对和近36万人工标注的音频-字幕对。该数据集专门设计用于风格标注文本到语音（CapTTS）、带声音效果的风格标注文本到语音（CapTTS-SE）、口音标注文本到语音（AccCapTTS）、情感标注文本到语音（EmoCapTTS）以及聊天机器人文本到语音合成（AgentTTS）任务。数据集包含音频路径、文本、来源、语音时长、音调、年龄、性别、语速、语音单调性、字幕、内在标签、情境标签、基本标签、所有标签、口音和背景噪声等信息。

CapSpeech is a benchmark dataset for style-captioned text-to-speech synthesis, containing over 10 million machine-annotated audio-caption pairs and nearly 0.36 million human-annotated audio-caption pairs. It is specifically designed for tasks such as style-captioned text-to-speech (CapTTS), style-captioned text-to-speech with sound effects (CapTTS-SE), accent-captioned text-to-speech (AccCapTTS), emotion-captioned text-to-speech (EmoCapTTS), and text-to-speech synthesis for chat agents (AgentTTS). The dataset includes information such as audio path, text, source, speech duration, pitch, age, gender, speaking rate, speech monotony, caption, intrinsic tags, situational tags, basic tags, all tags, accent, and noise.