ISCSLP 2026 CoT-TTS
A 16,000-hour bilingual context-aware CoT-guided TTS challenge dataset for Chinese and English
Duration
16000 hours
Languages
2
Sample Rate
various
Published
2026-06
Description
1Prepared for the ISCSLP 2026 CoT-TTS Challenge, targeting research on context-aware, expressive, and CoT-guided speech generation
2Contains approximately 16,000 hours and 3.0 million segments in total, including about 8,600 hours / 1.62 million English segments and 7,400 hours / 1.38 million Chinese segments
3Constructed from speech-rich media sources such as films, TV dramas, radio dramas, and short dramas, with rich dialogue context, speaker interactions, scene changes, and emotional variation
4Each sample is organized around a target utterance and provides historical dialogue context, reference speech, speaker labels, emotion tags, acoustic features, timestamps, and CoT-style analysis
5Audio is released in FLAC format; the official data card states that original acoustic properties are preserved as much as possible, so files may have different sampling rates, channel configurations, loudness levels, background sounds, and environmental noise
6A sampled FLAC file from movie-en1 dialogue_segments is 44.1 kHz, stereo, 16-bit, and 3.61 seconds long
7Released only for non-commercial academic research, development, and evaluation; commercial use, redistribution, resale, public re-hosting, repackaging, and attempts to reconstruct or identify original media sources are prohibited
Language Details
| Language | Duration |
|---|---|
| English | 8600 hours |
| Chinese | 7400 hours |
Publisher
License & Commercial Use
License
Non-commercial research-use license (following the spirit of CC BY-NC 4.0)
Commercial Use
Non-commercial only