Skip to content

Dataset · SparkAudio

VoxBox: Bilingual TTS corpus with metadata

SparkAudio/voxbox

Chinese and English speech data with clean transcriptions, age, gender, and emotion metadata for TTS.

What's inside

A collection of bilingual speech corpora for TTS in Chinese and English. It features clean transcriptions and rich metadata including age, gender, and emotion. Audio is organized into subcorpora with metadata stored in JSONL files (e.g., aishell-3, casia, commonvoice_cn, and wenetspeech4tts); the metadata size ranges from 10 to 100 million.

What you can do with it

  • Train a bilingual TTS model
  • TTS with emotion, age, and gender control
  • Evaluate speech synthesis in Chinese and English
Tasks
Text → speech
Languages
Chinese, English
Size
10M–100M rows
Downloads this month
5.8K
Likes
76

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("SparkAudio/voxbox")