Binary　metrics　is　employed　in　traditional　deep　learning　methods　of　sound　event　detection(SED)　to　determine　the　presence　or　absence　of　an　event.　However,　these　binary　activity　metrics　inadequately　characterize　the　nuanced　states　of　events,　which　limiting　the　performance　of　current　detection　algorithms,　particularly　in　scenarios　involving　event　overlaps.　Concurrently,　conventional　sound　event　detection　algorithms　suffer　from　sluggish　detection　speeds,　resulting　in　substantial　temporal　costs.　To　solve　the　above　problems,　a　novel　sound　event　detection　algorithm　based　on　amplitude　envelope　estimation　and　regression　detection(EERD)　is　proposed　in　this　paper.　In　this　algorithm,　firstly　the　envelope　of　the　Mel　Frequency　Cepstrum　Coefficient(MFCC)　of　the　audio　signal　is　estimatied,　thereby　enhanced　information　concerning　the　sound　events　is　obtained.　Secondly,　the　regression-based　detection　is　introduced　into　the　network　model,　so　that　the　algorithm＇s　reliance　on　post-processing　is　reduced　and　concomitantly　the　detection　speed　is　improved.　Empirical　validation　is　conducted　on　the　TUT　sound　event　detection　dataset.　Experiments　show　that　the　algorithm　proposed　in　this　paper　attains　superior　F-measure　for　sound　event　detection　contrast　to　benchmark　algorithms,　hence　heightened　detection　performance　is　substantiated.　At　the　same　time,　detection　speed　is　achieved　at　least　sixfold　faster　in　the　proposed　algorithm　than　the　conventional　segmentation-by-class　approach.　©　2023　IEEE.