The lab continues its work of creating, processing, analyzing and preserving corpora. A future project will identify all existing corpora, anonymize them and evaluate their metadata and the degree of confidentiality they require. In future, some corpora could be made available to all (for example, a corpus of aerodynamic data) while others could be available on a limited basis (for example, the Nakala corpus).
We continue to develop and use scripts and analysis protocols such as ema2wav, DataTranslationRecording, lippAnalyse, and MonPage. The LPP’s physiology system includes about fifteen pieces of equipment, and projects continue to take advantage of these devices, such as the AeroMask or the ePGG.
Several projects studying articulatory processes make good use of the recently-acquired electromagnetic articulograph (EMA, AG501). The use of EMA is extended to the collection of video data of respiratory movement, a novelty in research on respiration during speech that has not previously been reported. We are also looking into developing a procedure to acquire “minimal” articulatory data that would be suitable for studies requiring a very large number of speakers.
The use of computational methods implies continued collaboration with computer science labs (especially LIA and INRIA). Neural networks based on pre-trained unsupervised models are used to process prosodic and voice quality properties, as well as socially-conditioned variation. Methods that exploit orthographic transcription and the acoustic signal in parallel are also used.