Pranavz/emilia-en-mimi-q8-s4096-smoke-20260326e

---
pretty_name: emilia-en-mimi-q8-s4096-smoke-20260326e
language:
- en
task_categories:
- text-to-speech
size_categories:
- n<10K
---

# emilia-en-mimi-q8-s4096-smoke-20260326e

Frozen pretokenized Emilia-English model-ready dataset for TinyAya + Mimi training.

## Layout

- `train/lang=en/*.parquet`
- optional `validation/lang=en/*.parquet`
- optional `test/lang=en/*.parquet`
- `dataset_manifest.json`

## Selection

- source dataset: `amphion/Emilia-Dataset`
- data files: `Emilia/EN/EN-B000000.tar`
- source split: `train`
- quantizers: `8`
- train samples: `8`
- validation samples: `0`
- test samples: `0`
- min seconds: `1.0`
- max seconds: `30.0`

## Audio Codec

- backend: `mimi`
- source: `hf_pretrained`
- model: `kyutai/mimi`
- sample rate: `24000`

## Notes

This repo stores pretokenized training artifacts, not raw audio. Use `dataset_manifest.json`
as the immutable split fingerprint for ablation reproducibility.