Spanish audios classified according to phonetic-phonological speech disorders

The dataset was specifically constructed for the training and validation of a deep learning model, using voice recordings of preschool children with suspected phonetic-phonological disorders in Lima, Peru. For data collection, a web application was developed, designed to capture both the voice sample and its linguistic and clinical metadata in a structured manner. This tool would allow the therapist to manually enter the phonetic segment, the target phoneme, the phoneme's position in the word (initial, medial, or final), the type of articulatory error (omission, substitution, distortion, or correct production), and associate these attributes with a corresponding audio file. The audios could be recorded directly from the system using a web browser or later uploaded as external files.
The collection process took place in clinical and educational settings under the direct supervision of certified speech therapists. All recordings were captured using browsers on Android mobile devices, employing built-in microphones under controlled acoustic conditions. The audio files were stored in .wav format, with a sampling frequency of 16kHz and a resolution of 16 bits without compression.