rodenhhh/ContextTTS_dataset

---
license: cc-by-nc-4.0
task_categories:
- text-to-speech
- audio-to-audio

language:
- zh
tags:
- multi-modal
- tts-evaluation
- conversational-speech
size_categories:
- 1K<n<10K

configs:
- config_name: default
data_files:
- split: test
path: metadata.jsonl
default: true

---

# ContextTTS Evaluation Dataset

This is the official evaluation dataset for the paper **"[ContextTTS Eval: A Benchmark for Evaluating Long-Form
Contextual Expressive Text-to-Speech]"**. It is designed to evaluate the performance of multi-modal speech synthesis, specifically focusing on context-aware prosody and timbre consistency in Chinese conversations and audiobooks.

## Dataset Summary

The dataset consists of high-quality Chinese audio-text pairs, organized into three distinct categories to evaluate different aspects of TTS models:

1. **audiobook**: Long-form narrative speech with expressive prosody.
2. **conversation**: Multi-turn dialogues (e.g., from *Legend of the Demon Cat*) capturing natural interaction flows.
3. **timbre_prompt**: Reference audios used for zero-shot or few-shot timbre cloning evaluation.

## Data Structure

The files are organized as follows:
- `data/`: Contains sub-directories for each category.
- `metadata.jsonl`: The primary index file mapping audio files to their transcriptions and metadata.

### Data Fields
- `audio`: Path to the audio file (auto-loadable via `datasets` library).
- `transcription`: The corresponding text. For conversations, multiple turns are joined by newlines.
- `duration`: Audio duration in seconds.
- `category`: The source category (`audiobook`, `conversation`, or `timbre_prompt`).
- `dialogue_id`: Unique identifier for the conversation/session.

## Usage

### Preview on Hugging Face
You can use the **Dataset Viewer** tab on this page to listen to the samples and view the transcriptions directly in your browser. Use the **Filter** function on the `category` column to browse specific subsets.

### Programmatic Access
```python
from datasets import load_dataset

# Load the evaluation split
dataset = load_dataset("[Your-HF-Username]/[Your-Repo-Name]", split='test')

# Example: Filter for conversations only
conversations = dataset.filter(lambda x: x['category'] == 'conversation')
print(conversations[0])