A text-to-speech (TTS) evaluation set built from LibriSpeech test-clean (2620
utterances). Each example is a chat-style dialogue that asks a TTS model to read a
piece of text in a plain, neutral tone, paired with the ground-truth audio.
It is formatted for the ESPnet SpeechLM dialogue dataloader, but the schema is
generic and easy to consume from any framework.
.
├── dialogues.jsonl # 2620… See the full description on the dataset page:
https://huggingface.co/datasets/JinchuanTian/librispeech-test-clean-plain-tts.
