Software
Research software I have developed for my own work and for the people I work with. Each entry links to its GitHub repository, which holds the source code and the releases.
openphon
A phonetic analysis app for Android phones and tablets. It records or imports WAV files and shows the waveform, the spectrogram and the F0, intensity and formant tracks, aligned in time. Segmentation is done by touch on interval or point tiers and saved as a Praat TextGrid. For each interval, the duration, F0, intensity and the first three formants at the midpoint export to CSV, together with the analysis settings and the SHA-256 checksum of the source audio. Pitch-tracking errors can be corrected by hand, frame by frame. The exports and the voice report, which gives jitter, shimmer and HNR, then use the corrected values. All analysis runs on the device, in a Rust core that takes Praat's settings as its defaults. The measurements have been compared with Praat 6.1.38 on synthesized signals and on an evaluation set of 192 public Spanish recordings, where, at the tested settings, voicing decisions agree on 96.24% of frames and the median F0 difference is 0.32 Hz. These figures measure agreement with Praat under those conditions. An iOS version can be built from the source but is not distributed.
Release candidate 0.1.0-rc.5Apache-2.0
Marcadores de Paula
A desktop application for Windows and Apple-silicon Macs that transcribes recordings and lets the result be checked and annotated. It downloads audio or video from a link or opens local files, transcribes them on the user's own computer with WhisperX (Whisper large-v3 or large-v3-turbo), aligns every word to the audio and separates speakers with pyannote. In the review editor each turn can be played on its own to correct its text, its speaker or its boundaries. The original machine transcript is kept beside the reviewed one. The application was made for research on discourse markers in spoken Spanish, so it applies no cleanup of its own to the repetitions, filled pauses or false starts the recogniser keeps. An annotation layer kept apart from the text marks them, with an editable codebook of 27 labels and 21 functions that draws on the Val.Es.Co. conventions. The transcript exports to several formats, Praat TextGrid and ELAN EAF among them.
Version 0.9.1GPL-3.0-or-later
ATALAYA
An Android app for language portraits: each participant colours a body silhouette with the languages in their life and explains where each one goes and why. ATALAYA records that explanation while the portrait is drawn and transcribes it on the phone with Whisper small, without sending the audio to any server. The drawing, the audio, the transcript and the participant's details of each portrait are stored under a participant code and can be shared from the app. It was developed in the linGAP research group at the Universidad de Salamanca by María Carmen Fernández Juncal, Olga Ivanova, Pedro Álvarez Mosquera and me; I did the programming. It is registered in the Registro Central de la Propiedad Intelectual (entry 00/2026/5495, effective 30 June 2026), with the Universidad de Salamanca as holder of the exploitation rights.
Version 1.7.4, Android 7.0 or later