---
language:
- en
tags:
- text-to-speech
- tts
- nonverbal
- fluac
- codec-tokens
license: cc-by-nc-sa-4.0
---

# VUI English NonverbalTTS FLUAC Q9 22kHz v1

English NonverbalTTS subset prepared for VUI decoder distillation.

## Processing

- Source dataset: `deepvk/NonverbalTTS`
- Text source: validated `Result` / fallback text
- Audio codec: FLUAC Q9 22kHz
- `fluac_codes` shape: `[9, T]`
- No EOS/PAD frames included
- Training collator should add EOS if needed
- Final `[pause]` is appended to match VUI-style prompts
- Unsupported emoji/nonverbal events are dropped
- Supported mappings:
  - `🌬` -> `[breath]`
  - `🤣` -> `[laugh]`
  - `😤`, `😖` -> `[sigh]`
- Unsupported dropped examples include cough/sniff/sneeze/snore/grunt-like events.

## Stats

- Records: 2444
- Duration: 6.38 hours
- Splits: train 2066, valid 343, test 35
- Max frames: 643
- Emoji rows: 0
- Prosody counts:
  - `[breath]`: 4210
  - `[sigh]`: 1088
  - `[laugh]`: 294

## Main columns

- `text_train_nograph`
- `text_train_default`
- `conditioned_text`
- `text`
- `fluac_codes`
- `duration_sec`
- `num_frames`
- `split`
- `speaker`
- `emotion`