aether-raid/atc-tts-voxtream

## ATC TTS Voxtream Dataset

This dataset is prepared for training the Voxtream TTS model and follows the format of [herimor/voxtream-train-9k](https://huggingface.co/datasets/herimor/voxtream-train-9k).

**Source:**
Based on the Singaporean SG Eleven dataset: [aether-raid/sg-aviation-el-combined](https://huggingface.co/datasets/aether-raid/sg-aviation-el-combined)

**Description:**
- All features are provided as single batched `.npy` files:
- `mimi_codes_16cb.npy`: Mimi codec tokens (16 codebooks)
- `phone_emb_indices.npy`: Alignment of phoneme tokens to Mimi frames
- `phone_tokens.npy`: Phoneme tokens
- `sem_label_shifts.npy`: Monotonic phoneme alignment labels
- `spk_templates.npy`: Speaker templates (first 3 seconds, if available)
- `uids.npy` contains the list of utterance IDs, matching the batch dimension of all features.

**Usage:**
See [herimor/voxtream-train-9k](https://huggingface.co/datasets/herimor/voxtream-train-9k) for loading and training examples.

**License:**
See original dataset for licensing and usage restrictions.