Closed ZacharyYarostTechsmith closed 1 month ago
This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 7 days.
I've yet to find the shift between DFN2 and DFN3 that leads to this behavior.
This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 7 days.
Hello! Firstly I'd like to thank you for your work building this library and model. It's an absolutely fantastic project.
While testing DFN I've found a scenario in which version 3 silences human speech, even in a mostly noise free environment. Here is a video comparing the outputs of version 1, 2, and 3 with one such example.
I'd be grateful to get your feedback on this behavior. I question if it's simply filtering out audio below a certain volume threshold.
I've attached the associated example clips below.
LowVolumeSpeakerTesting_Mono_48kHz_16Bit.zip