Add Audio8-TTS-Realistic voice dataset

README.md

@@ -0,0 +1,49 @@
+---
+license: apache-2.0
+task_categories:
+- text-to-speech
+language:
+- en
+tags:
+- synthetic
+- tts
+- audio
+size_categories:
+- n<1K
+---
+
+# Audio8 TTS Realistic - synthetic voice dataset
+
+Synthetic speech generated with [Edge0/Audio8-TTS-Realistic-0.6b](https://huggingface.co/Edge0/Audio8-TTS-Realistic-0.6b).
+
+- **Clips:** 11
+- **Total audio:** 0.03 hours
+- **Sample rate:** 44100 Hz, mono, 16-bit
+- **Clip length:** 7-18 seconds
+- **Language:** English
+- **Voice:** All clips use one cloned reference voice.
+- **Text:** emotional lines written by a small LLM (Qwen/Qwen3.8-Flash-Next) unless a custom text file was supplied; the emotion is carried by wording and punctuation
+
+## Columns
+
+| column | description |
+|---|---|
+| `id` | unique clip id |
+| `audio` | the waveform (embedded, decoded by `datasets`) |
+| `text` | the spoken text |
+| `emotion` | the emotion the text was written to express (empty for custom texts) |
+| `language` | language name |
+| `duration` | length in seconds |
+
+## Usage
+
+```python
+from datasets import load_dataset
+ds = load_dataset("Shadow0482/tts-realistic", split="train")
+print(ds[0]["text"], ds[0]["audio"]["sampling_rate"])
+```
+
+## Notes
+
+Fully synthetic audio. Clips were filtered automatically (silence, NaNs, runaway generations) but not
+listened to individually, so expect occasional pronunciation errors. Please disclose synthetic audio when you use it.

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d03d97275d9e4ff8545cc5dad93fb0535321a0910205414f3d5e8544d31494b8
+size 8228583