Rich Voice Dataset for Emotion Recognition & Speech AI

**Overview**
Emotion understanding is a crucial pillar for human-centered AI. This dataset provides 50,000 synthesized, emotion-annotated voice recordings from 500 real speaker voices, offering a high-quality audio benchmark for detecting and modeling emotional speech. It supports diverse applications such as voice assistants, mental health monitoring, empathetic AI, and voice with emotional nuance.

**Use cases**

- Emotion Recognition Training: Build or evaluate speech models that can classify emotional tone with real-world variability.
-
- Expressive Text-to-Speech (TTS): Train voice synthesis models to generate speech with expressive emotional cues.
-
- Mental Health & Wellness Applications: Detect signs of distress or emotional shifts through vocal cues in real-time.
-
- Accessibility & Social Robotics: Enable more emotionally aware human-computer interactions.

**Product details**

- Metadata CSV: Metadata and annotations per speaker and line.
-
- Audio_files/: Folder containing all WAV files.
-
- Speaker_Profiles: Included in the metadata csv.

Sample Fields:

text: Emotionally rich sentence

emotion_label: Primary labeled emotion (e.g., joy, anger, fear)

audio_file_path: Path to the voice file

gender, age_range, region, native_language: Speaker metadata

For more details, refer to the embedded notebook.

**Additional Insights**
Inspired by the DAIR Twitter Emotions dataset, this voice corpus bridges the gap between text-based emotion classification and real-world speech-based emotion understanding. The data was ethically collected and fully consented, ensuring responsible AI development. The balance across emotions and speaker demographics enables equitable performance across diverse populations.

For more details, please feel free to reach out to us at [e-mail]

搜集汇总

背景与挑战

背景概述

该数据集包含50,000条来自500位真实说话者的合成语音记录，带有情感标注，适用于语音情感识别、表达性语音合成等AI应用。数据包含音频文件、元数据及说话者人口统计信息，覆盖美加地区，并遵循伦理收集标准。