# Reference voice attributions

The reference clips in this project are used as short conditioning samples for multilingual speech synthesis. Each published episode remains original project content.

## English LibriSpeech reference

The English reference clip is derived from LibriSpeech speaker 5338, which in turn is based on a LibriVox audiobook recording.

- Dataset: https://huggingface.co/datasets/openslr/librispeech_asr
- Project: https://www.openslr.org/12/
- License: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/
- Exact split, row, transcript, ASR match, voiced duration, and checksum are recorded in `ref_voice/15_Multilingual_Male/manifest.json`.

## Mozilla Common Voice references

Japanese and French reference clips are derived from Mozilla Common Voice data released under CC0 1.0.

- Source: https://commonvoice.mozilla.org/cv/datasets
- License: https://creativecommons.org/publicdomain/zero/1.0/
- Exact dataset, split, row, transcript, and checksum details are recorded in `ref_voice/15_Multilingual_Male/manifest.json`.

## Zeroth Korean

The Korean reference clip is derived from the Zeroth Korean speech corpus through the `Bingsu/zeroth-korean` Hugging Face mirror.

- Dataset: https://huggingface.co/datasets/Bingsu/zeroth-korean
- License: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/

## Multilingual LibriSpeech

German, Portuguese, Spanish, and Italian reference clips are derived from Multilingual LibriSpeech by Meta AI, sourced from LibriVox audiobook readings.

- Dataset: https://huggingface.co/datasets/facebook/multilingual_librispeech
- License: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/
- Project: https://www.openslr.org/94/

## CSS10 / LibriVox Russian reference

The Russian reference clip is derived from the CSS10 multilingual speech collection and its public-domain LibriVox source reading, using the `davidguzmanr/CSS10-Multilingual-LJSpeech` dataset mirror.

- Dataset mirror: https://huggingface.co/datasets/davidguzmanr/CSS10-Multilingual-LJSpeech
- Mirror license: Apache 2.0 — https://www.apache.org/licenses/LICENSE-2.0
- CSS10 paper: https://arxiv.org/abs/1903.11269

No reference clip is presented as an endorsement by its original speaker or dataset publisher.