Respiratory signal extension
The specific technical contribution NIPG would layer on top of the Szeged-origin Speech Gap IP, and the clearest example of how the ecosystem’s parts fit together.
The limitation
The current Speech Gap software treats respiratory pauses as silence. For the dementia and mild-cognitive-impairment population the method was built for, this is acceptable. For BPD it may not be: breathing changes, including possible heavy or rapid breathing, may themselves carry relevant information, and collapsing them into “silence” discards that signal entirely.
The proposed extension
Distinguish six interval types instead of two:
- Silence — true absence of both speech and breath sound
- Inhalation
- Exhalation
- Filled pauses — vocalised hesitation
- Speech
- Uncertain intervals — explicitly labelled rather than silently assigned
Then test whether the resulting respiratory features are reliable (do they measure consistently?) and clinically informative (do they relate to anything that matters?). Both questions are open.
Why this is a natural NIPG contribution
It is signal processing and feature engineering over an audio stream — squarely within the capability portfolio — and it does not touch the patented method itself, only what is extracted from the same recordings. It also illustrates the intended division of labour: Szeged owns the origin IP and the validated instrument; NIPG extends the measurement layer.
Dependencies
- The extension changes what the measurement produces. Under PCI Registered Reports discipline, this must be specified, validated and frozen before it is used on outcome-relevant study data.
- It requires an annotation scheme and inter-annotator agreement for the new interval types. → Data governance and annotation
Related pages
Speech Gap · Temporal speech parameters · NIPG · PCI Registered Reports · Data governance and annotation