GianDiego/latam-spanish-speech-orpheus-tts-24khz

LATAM Spanish High-Quality Speech Dataset (24kHz - Orpheus TTS Ready)

Dataset Description

This dataset contains approximately 24 hours of high-quality speech audio in Latin American Spanish, specifically prepared for Text-to-Speech (TTS) applications like OrpheusTTS, which require a 24kHz sampling rate.

The audio files are derived from the Crowdsourced high-quality speech datasets made by Google and were obtained via OpenSLR. The original recordings were high-quality (48kHz, 16-bit, mono) and carefully trimmed to remove leading/trailing silences.

This version has been processed:

• Resampled: Audio converted from 48kHz to 24kHz.
• Format: Saved as 16-bit PCM WAV mono.
• Structured: Organized into a Hugging Face datasets format with metadata.

The dataset features speakers from various Latin American countries, providing a range of accents and voice timbres:

• Argentina (ar)
• Chile (cl)
• Peru (pe)
• Colombia (co)
• Venezuela (ve)
• Puerto Rico (pr)

Each audio file has a corresponding transcription.

How to Use

You can load the dataset using the datasets library:

from datasets import load_dataset

repo_name = "GianDiego/latam-spanish-speech-orpheus-tts-24khz"
dataset = load_dataset(repo_name, split="train")

sample = dataset[0]
print(sample["audio"])
print(sample["text"])
print(f"Nationality: {sample['nationality']}, Gender: {sample['gender']}, Speaker ID: {sample['speaker_id']}")

Dataset Structure

Data Fields

• audio: A dictionary containing the path to the audio file, the decoded audio array, and the sampling rate (fixed at 24kHz).
• text: The transcription corresponding to the audio file.
• file_id: The unique identifier derived from the original filename (e.g., vem_05223_00896110924).
• nationality: The 2-letter country code indicating the speaker's origin (ar, cl, pe, co, ve, pr).
• gender: The speaker's gender (m for male, f for female).
• speaker_id: The identifier for the specific speaker within their nationality/gender group (e.g., 05223).

Data Splits

The dataset contains a single split: train.

Dataset Creation

Source Data

The original audio data comes from Google's Crowdsourced high-quality speech datasets, distributed via OpenSLR:

• SLR71 (Chile)
• SLR73 (Peru)
• SLR86 (Colombia)
• SLR72 (Venezuela)
• SLR74 (Puerto Rico)
• SLR85 (Argentina) (Users should verify the exact source datasets used if needed)

These datasets were originally recorded at 48kHz, 16-bit PCM mono and are described in detail in the paper cited below.

Annotations

The transcriptions were provided alongside the original audio files by Google/OpenSLR. They generally represent the spoken content accurately. Some minor variations or specific notations (like a_letter observed in some examples) might be present from the original source. This processed dataset retains the original transcriptions.

Personal and Sensitive Information

The dataset consists of spoken utterances collected via crowdsourcing. While efforts are usually made to anonymize data (e.g., using anonymized FileIDs), the nature of speech means it could potentially contain identifiable characteristics or incidental mentions. Users should use the dataset responsibly and in accordance with the license terms.

Licensing Information

The original datasets from OpenSLR are licensed under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) License.

In accordance with the ShareAlike clause, this derived dataset (resampled to 24kHz and formatted for Hugging Face) is also distributed under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) License.

This means you are free to:

• Share — copy and redistribute the material in any medium or format
• Adapt — remix, transform, and build upon the material for any purpose, even commercially.

Under the following terms:

• Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
• ShareAlike — If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.

Please review the full terms of the license: https://creativecommons.org/licenses/by-sa/4.0/

Citation Information

If you use this dataset in your work, please cite both the original data collection effort and this specific processed version:

1. Original Data Collection (Google/OpenSLR):

@inproceedings{guevara-rukoz-etal-2020-crowdsourcing,
    title = {{Crowdsourcing Latin American Spanish for Low-Resource Text-to-Speech}},
    author = {Guevara-Rukoz, Adriana and Demirsahin, Isin and He, Fei and Chu, Shan-Hui Cathy and Sarin, Supheakmungkol and Pipatsrisawat, Knot and Gutkin, Alexander and Butryna, Alena and Kjartansson, Oddur},
    booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},
    year = {2020},
    month = may,
    address = {Marseille, France},
    publisher = {European Language Resources Association (ELRA)},
    url = {https://www.aclweb.org/anthology/2020.lrec-1.801},
    pages = {6504--6513},
    ISBN = {979-10-95546-34-4},
}

2. This Processed 24kHz Version:

@misc{latam_speech_24khz_giandiego_2024,
  author       = {GianDiego and Google Contributors}, # Giving credit as per license
  title        = {LATAM Spanish High-Quality Speech Dataset (24kHz, Orpheus TTS Ready)},
  year         = {2024}, # Or the year you upload it
  publisher    = {Hugging Face},
  journal      = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/datasets/GianDiego/latam-spanish-speech-orpheus-tts-24khz}}
}

Acknowledgements

Thanks to Google for creating and releasing the original high-quality speech data, the authors of the LREC paper for documenting the crowdsourcing effort, and to OpenSLR for hosting the resources.