Ideal binary masks identify target-dominant time-frequency tiles for glimpse resynthesis
Evidence
To generate the glimpsed stimuli, an ideal binary mask was applied separately to the two ears of the binaural stimuli using the methods of Wang ( 2005 ) and Brungart et al. ( 2006 ). In short, the signals were analyzed using 128 frequency channels between 80 and 8000 Hz, and 20-ms time windows with 50 % overlap. Tiles with target energy exceeding masker energy were assigned a mask value of one and the remaining tiles were assigned a value of zero. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1753 (
#corroborates, via linked claim)
Categorization
- XREF: connects to: Ideal binary masking isolates time-frequency tiles where target dominates masker — Dijk, Pim van, Physiology, Psychoacoustics and…, 1753
Related
- Ideal binary masking isolates time-frequency tiles where target dominates masker (similarity 0.96)
- Ideal binary masks isolate glimpses where target exceeds masker energy. (similarity 0.93)
- Glimpsing models quantify target audibility by time-frequency tiles where target exceeds masker (similarity 0.86)
- Speech perception depends on brief time-frequency glimpses where target energy exceeds masker energy. (similarity 0.85)
- Target audibility in speech mixtures requires measuring glimpsed time-frequency regions, not quiet thresholds (similarity 0.85)
- Binary spectral masking selects target components based on local SNR thresholds (similarity 0.82)
- Quiet-only performance measures miss masking effects that reduce speech redundancy. (similarity 0.79)
- Binaural unmasking depends on masker duration but not signal duration (similarity 0.78)
- Each masker dominates masking only within its own frequency region. (similarity 0.78)
- Separating signal from masker locations improves detection through two binaural effects (similarity 0.78)
- Spatial release from masking persists even when target speech components are selectively discarded (similarity 0.78)
Broader Connections
- Ideal binary masking isolates time-frequency tiles where target dominates masker
#corroborates(0.98) — Dijk, Pim van, Physiology, Psychoacoustics and…, 1753