LaFresCat: A Catalan Multi-Accent Speech Dataset for Text-to-Speech

<h1>LaFresCat Multiaccent</h1>\n<p>We present LaFresCat, the first Catalan multiaccented and multispeaker dataset.</p>\n<p>This work is licensed under a <a href="https://creativecommons.org/licenses/by-nc/4.0/deed.en" rel="nofollow">Creative Commons Attribution-NonCommercial 4.0 International License</a>. Commercial use is only possible through licensing by the voice artists. For further information, contact <a href="mailto:[email protected]" rel="nofollow">[email protected]</a> and <a href="mailto:[email protected]" rel="nofollow">[email protected]</a>.</p>\n<h2><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#dataset-details" rel="nofollow"> </a></h2>\n<h2>Dataset Details</h2>\n<h3><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#dataset-description" rel="nofollow"> </a></h3>\n<h3>Dataset Description</h3>\n<p>The audios from this dataset have been created with professional studio recordings by professional voice actors in Lafresca Creative Studio. This is the raw version of the dataset, no resampling or trimming has been applied to the audios. Audios are stored in wav format at 48khz sampling rate</p>\n<p>In total, there are 4 different accents, with 2 speakers per accent (female and male). After trimming, accumulates a total of 3,75h (divided by speaker IDs) as follows:</p>\n<ul>\n<li>\n<p>Balear</p>\n<ul>\n<li>olga -> 23.5 min</li>\n<li>quim -> 30.93 min</li>\n</ul>\n</li>\n<li>\n<p>Central</p>\n<ul>\n<li>elia -> 33.14 min</li>\n<li>grau -> 37,86 min</li>\n</ul>\n</li>\n<li>\n<p>Occidental (North-Western)</p>\n<ul>\n<li>emma -> 28,67 min</li>\n<li>pere -> 25,12 min</li>\n</ul>\n</li>\n<li>\n<p>Valencia</p>\n<ul>\n<li>gina -> 22,25 min</li>\n<li>lluc -> 23,58 min</li>\n</ul>\n</li>\n</ul>\n<h2><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#uses" rel="nofollow"> </a></h2>\n<h2>Uses</h2>\n<p>The purpose of this dataset is mainly for training text-to-speech and automatic speech recognition models in Catalan dialects.</p>\n<h3>Languages</h3>\n<p>The dataset is in Catalan (<code>ca-ES</code>).</p>\n<h2>Dataset Structure</h2>\n<p>The dataset consists of 2858 audios and transcriptions in the following structure:<code><br><br>lafresca_multiaccent_raw<br>├── balear<br>│   ├── olga<br>│   ├── olga.txt<br>│   ├── quim<br>│   └── quim.txt<br>├── central<br>│   ├── elia<br>│   ├── elia.txt<br>│   ├── grau<br>│   └── grau.txt<br>├── full_filelist.txt<br>├── occidental<br>│   ├── emma<br>│   ├── emma.txt<br>│   ├── pere<br>│   └── pere.txt<br>└── valencia<br>    ├── gina<br>    ├── gina.txt<br>    ├── lluc<br>   └── lluc.txt</code></p>\n<p>Metadata of the dataset can be found in the file `full_filelist.txt` , each line represents an audio and follows the format:</p>\n<p>audio_path | speaker_id | transcription</p>\n<p>The speaker ids have the following mapping: </p>\n<p>"quim": 0,<br>"olga": 1,<br>"grau": 2,<br>"elia": 3,<br>"pere": 4,<br>"emma": 5,<br>"lluc": 6,<br>"gina": 7</p>\n<h2>Dataset Creation</h2>\n<p>This dataset has been created by members of the Language Technologies unit from the Life Sciences department of the Barcelona Supercomputing Center, except the valencian sentences which were created with the support of Cenid, the Digital Intelligence Center of the University of Alicante. The voices belong to professional voice actors and they've been recorded in Lafresca Creative Studio.</p>\n<h3><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#source-data" rel="nofollow"> </a></h3>\n<h3>Source Data</h3>\n<p>The data presented in this dataset is the source data.</p>\n<h4><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#data-collection-and-processing" rel="nofollow"> </a></h4>\n<h4>Data Collection and Processing</h4>\n<p>These are the technical details of the data collection and processing:</p>\n<ul>\n<li>\n<p>Microphone: Austrian Audio oc818</p>\n</li>\n<li>\n<p>Preamp: Focusrite ISA Two</p>\n</li>\n<li>\n<p>Audio Interface: Antelope Orion 32+</p>\n</li>\n<li>\n<p>DAW: ProTools 2023.6.0</p>\n</li>\n</ul>\n<p>Processing:</p>\n<ul>\n<li>\n<p>Noise Gate: C1 Gate</p>\n</li>\n<li>\n<p>Compression BF-76</p>\n</li>\n<li>\n<p>De-Esser Renaissance</p>\n</li>\n<li>\n<p>EQ Maag EQ2</p>\n</li>\n<li>\n<p>EQ FabFilter Pro-Q3</p>\n</li>\n<li>\n<p>Limiter: L1 Ultramaximizer</p>\n</li>\n</ul>\n<p>Here's the information about the speakers:</p>\n<div>\n<table>\n<tbody>\n<tr>\n<th>Dialect</th>\n<th>Gender</th>\n<th>County</th>\n</tr>\n</tbody>\n<tbody>\n<tr>\n<td>Central</td>\n<td>male</td>\n<td>Barcelonès</td>\n</tr>\n<tr>\n<td>Central</td>\n<td>female</td>\n<td>Barcelonès</td>\n</tr>\n<tr>\n<td>Balear</td>\n<td>female</td>\n<td>Pla de Mallorca</td>\n</tr>\n<tr>\n<td>Balear</td>\n<td>male</td>\n<td>Llevant</td>\n</tr>\n<tr>\n<td>Occidental</td>\n<td>male</td>\n<td>Baix Ebre</td>\n</tr>\n<tr>\n<td>Occidental</td>\n<td>female</td>\n<td>Baix Ebre</td>\n</tr>\n<tr>\n<td>Valencian</td>\n<td>female</td>\n<td>Ribera Alta</td>\n</tr>\n<tr>\n<td>Valencian</td>\n<td>male</td>\n<td>La Plana Baixa</td>\n</tr>\n</tbody>\n</table>\n</div>\n<h4><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#who-are-the-source-data-producers" rel="nofollow"> </a></h4>\n<h4>Who are the source data producers?</h4>\n<p>The Language Technologies team from the Life Sciences department at the Barcelona Supercomputing Center developed this dataset. It features recordings by professional voice actors made at Lafresca Creative Studio.</p>\n<h3><a href="https://huggingface.co/datasets/projecte-aina/LaFrescat/blob/main/README.md#annotations" rel="nofollow"> &