This is a corpus of the Silbo Gomero whistled language, which is a whistled form of Spanish used on the La Gomera island. It was created from 49 minutes of raw recordings. The recordings contained read speech, and were produced by 4 fluent whistlers. They were created for use in teaching this language to children native to the island.
<p>
The corpus consists of 3 parts, each of which was made from the same data, edited in different ways; separate transcription file is provided for each part.
<ul>
<li>'words.zip' contains clips of single, separate words. Some clips may contain more than one word, in cases where the separation was not possible.</li>
<li>'sentences.zip' contains clips of entire sentences. Some parts of the recordings are not represented here; for example, one recording contained a poem, which could not be separated into sentences.</li>
<li>'fragments.zip' contains clips of short fragments of speech (on average, about 6.5 words long); those fragments were made by separating recordings where longer pauses between words occured.</li>
</ul>
<p>
This corpus was created by Agata Jakubiak, a student at University of Warsaw, from data provided by Francisco Javier Correa, working for the Silbo Gomero Teaching Project (Proyecto de Enseñanza de Silbo Gomero), as a part of research into Automatic Speech Recognition of whistled speech.
<p>
You can cite the data using the following BibTeX entry:
<pre>
@inproceedings{jakubiak23_interspeech,
  author={Agata Jakubiak},
  title={{Whistle-to-text: Automatic recognition of the Silbo Gomero whistled language}},
  year=2023,
  booktitle={Proc. INTERSPEECH 2023},
  pages={3402--3406},
  doi={10.21437/Interspeech.2023-989}
}
</pre>