Lack of "capture" functionality in MSpectralDynamicsLE Denoiser. Any impact for removing vocal background noise?

[carloscadu]

Hi,

I'd appreciate your feedback.
Since the Denoiser device in MSpectralDynamicsLE can't "capture" the noise profile like in the full MSpectralDynamics, should it be a concern for performing at full potential the denoising of removing vocal background noise?

What would be the impacts of not having the "capture" button in MSpectralDynamicsLE to learn the noise profile?

The noise I'm referring to is continuous ambient noise with a stronger presence of computer's fan. Sometimes air-conditioner noise is present too.

Thank you,
Carlos

[MeldaProduction]

Hi Carlos, thanks for the feedback. I'm afraid while I technically wouldn't mind having that feature, it's quite problematic to implement at this moment (long story). But I'll check again.

[carloscadu]

Thanks for replying Vojtech.

Given the default controls on MSpectralDynamicsLE Denoiser device, I only change "threshold" to match the peak of the noise I want to get rid of. For instance, I usually set the threshold to -55dB.

My concern is, once there isn't noise profile capture functionality, and the noise removal is based mainly on the "threshold", is there any risk that MSpectralDynamicsLE can remove other content than noise from the voice signal?
Could you please confirm the threshold control is pretty much what needs to be set and MSpectralDynamicsLE Denoiser will not affect the voice signal once the threshold is set properly?

[CableChannel]

I have programmed a denoiser for vocals preset that works fantastically. I'll post it tomorrow, maybe it works for you too.

[carloscadu]

I have programmed a denoiser for vocals preset that works fantastically. I'll post it tomorrow, maybe it works for you too.

Wonderful. It's very kind of you CableChannel!

[CableChannel]

Try these. They work pretty well for me on speech as long as the noise isn't too loud.
Most important to tune them is threshold and ratio (for which I made multiparameters called threshold and strength). Basically that works analogous to a "compressor knee" for the denoiser, you can have e.g. a high threshold with high ratio meaning the denoiser kicks in late, but then very strongly. Or the other way around, low threshold and low ratio, meaning the denoiser kicks in early, but rather mildly.

If you can post / send by PM a speech sample I can also try to tune the values for you.

Code: Select all

$eNq9Vt9vGjkQfuevsIbXBmzvsj+kXaqU0CpSSFE2l3v2wQSsGJuzvW3IQ--203qBAEna9E66fYGZHX-z+ZthhuLj40qRb2idNLoE1qNAUM-MXOpFCbW-P8vg47BTTKo1zrwV6mKjxUrOnEPvpV440i2BAemeey9mDyVQIN2F8PigEZ8tv7TolkbNS6A9mlDO0jxjbMBzIN0m1MknZG382poZOteAr4zxy3CE0-bheQak+3ctlPSbEiY4l-UKSNcKLw0rIeoN8u2TpEC6N6hQOAwYtOFp0RlV+3Bb2qOMsjgepCzmadKET6or1IuQFEi3Woo1TswcS7iSGoUF0j0gRek2VZomccNrf83mKj2eRFEW5WkeMSCj2lrUfmrRob+8KOF2aazzqH-cfZ0JRS5QG+nQEgZkPJfe2Eo+YQksSQcfEhYdesfCbUrIWf5hQHMg48e10HO0LFTiSjh-UIzGfBahtW+fq3HGox5Lk5gPMjoAciX1w2gptEblWjRjHs6XKOYXqMSmRbip1kp6j4FLk5Rm7NTNS+BpDmRSKy+nwooVerRuZJQSa4fzgLTl1Qr8WTgPpFqa7+NQ3ie0X6xYt5Wo5BxHSyH1dMdi76lwZvRc2M2Wd3h7t2vonOU8T2BYjIz21iiF1h0alIy1+Es1hBiQPdFrscISgkkY9A9PMPL8vQTelOs3EfgxAn8HAj9GiI4RoncgRMcI8TFC-A6EuEHoH+rYKU7KOzxx0Jdtf6nXtScLITWQC7wXtfJ3QtXhF5oyxlMeD6J4AOTLH5dtxzftwUu4NnYlFJC2W3ZWS3Df0ECqNW676xQ2zRiNt+4W5NOG3EvrPFnvCMOw2JMnn2ulboReYAnxS03aKWUsYeSMHBDYpQ1qnchxqg97qc-X2r8tUBbRjMeM5Tnjv6dQ5W2Ya68KtIWNkjT6DwJNxOMeMc-zZlwznqWDX0h300zun8rGXnia1jufefkNW+F2Mh76LvW9KQFee9XS2E3cdrI2fd9vOjrg82Hxp5z75WRYVKa2MyRTi-dSebRhMbY0dyH93aFOsWhG1rCYGqk9+azEwn0qgcVhApw6KZDHsEdimqRRmuZZFvGfBR4tlM2pp-9a2uZgGxwC+i2-TjGqnTersN9cc+39zCXj67vpzbga3zaDPUiyb+4wj8mFFd+lXrQdwuDNrL2IMsrzPB8kSWBwvGvepBv-+mD-mNKwU4TPdu39O4LvpNMkP8i1y7zdZP9P6m2yYafov-W3zA07-wBnWxEC

Code: Select all

$eNq9Vk1z4jgQvfMrVM11ApZsjF1lM5UBMpWqIUvhbPasxR1QRUiMJM+EHPa3b1nGhI-JbGYO6wuo3Xr9+qnV7ezj80aSb2is0CoH2guAoFrqUqhVDpV7vErg46iTzYotLp3hcrJTfCOW1qJzQq0s6eZAgXSvnePLpxwCIN0Vd-ikEF9Xbm3QrrUscwh6QRwwOkwTSgcsBdKtXa14Qdr4b41eorU1+EZrt663UBYlQRAEbDAcAOl+rbgUbpfDDEtRbYB0DXdC0xxYL22fJAbSXaBEbtGHDWqeBq2WlfPZtrbFrPiCauVDAekWa77FmS4xh+JrxQ2WQLoHLkC6h2xqxr1wECU0DJOUDiIg48oYVG5u0KK7neRwv9bGOlT-POgll2SCSguLhjAg01I4bQrxgjnQeDj4ENPw2DrldpdDStMPgyAFMn3eclWioV7xL9y6I9Hr5Wuyzfr+VfUrmvSC-ROmQL4I9TRec6VQ2gZO66frNfJygpLvGohFsZXCOfRk6qhBQs-NLAc2TIHMKunEnBu+QYfGjrWUfGux9Eh7Yo2mN9w6IMVaf5-6c3xB89nwbaNtIUocr7lQ85bFwVLgUquSm92et3-70FZuSlOWxjDKxlo5o6VEY48XAZkq-resCVEgB6J3fIM5+CWh0D-eQcnr-xwYhV9GYKcI7B0I7BQhPEUI34EQniJEpwjROxCiGqF-rGMnOzve0ZkhuKz7W7WtHFlxoYBM8JFX0j1wWfmrOIwTxlgSp1HIgHz+87ap+bo+WA532my4BNKUS7tqGB5KGkixxX15edzG+9OOPApjHdm21GCUHWiSm0rKBVcrzCG6zL5pPNoQSq7IUaSWt9flLPFzJeilEn9U7m0p0pTSOBrGQRTSX5OicMY3rQsljmCj+q7-rjwz-vx6YgNWN480jOL-0G1Rd2Ige5iznT9Rkl5Y6rq7XjrxDRstW2WPbbfqUecAP3rVkGv7bdNX66Lv1+Xs8dko+0uUbj0bZYWuzBLJ3OCjkA6NH38Nzdal327qZKu6X42yuRbKkRvJV-ZTDjTy1--cGAB59qMmjGg0iJJkyNLBzxxP58nuwtT-UeB6K-XenkS-YdjJxpV1euMnmq0TP7RcMr17mC+mxfS+7utelEPF+3ZMJoZ-F2rVFA6FN6P2woAGLE3TQRx7Bmez5k2+0Tt29k9JjTqZ-20G3+9RfCefOvhRrDbyfpT9P6H3wUadrP-WB5gddf4Fp4oMhQ==

Code: Select all

$eNq9Vl1vKjcQfedXWOb1Bmzvt7TLVS6Qq0ghRdk0fXZhAlaMTW3vTchDf3tlL99J2tyqKi-g2ZkzZ45nZyi-vqwk+gHGCq0qTHsEI1AzPRdqUeHGPV7k+OugU07qNcyc4XK0UXwlZtaCc0ItLOpWmGLUvXSOz54qTDDqLriDJwVwOLmlAbvUcl5h0iMpYTQrckoTVmDU9a5WvAJt-ddGz8BaD77S2i1DCCXth8Y+4o+GS+E2Fb7Rzxh1DXdC0wpHPVYkcV4UBSE5Rt07kMAtBADiSRqwWjYulBpAaRwnGY1Zlnr3SX0DahEyYtStl3wNEz2HCt8IBdxg1D0w8nD7qjzzXhRFUZ4xVtC8wGjYGAPKTQ1YcNejCt8vtbEO1J8PesYlGoHSwoJBEUbjuXDa1OIVKkzTLPmS0hPrmNtNhQtafElIgdH4Zc3VHAwNyt9w647E98dD3e35-qD+BS16JGckpilNfLRQT8MlVwqkbeG0frpcAp+PQPJNC3FXr6VwDgIZn5Xk9NzMKsyyAqNJI52YcsNX4MDYoZaSry3MA9KWWCvqFbcOo3qpn8fhPl-BfDd83apfizkMl1yo6Y7F3lLDTKs5N5st7-D0YdfBBS1YkeJBOdTKGS0lGHt8IGis+O-SE6IY7Yne8hVUOBwRxf3jCIoOvyvMKP5pBHaKwD6BwE4RolOE6BMI0SlCfIoQfwIh9gj9Yx075dn1Ds4M5G3fX6t149CCC4XRCB55I90Dl014K7MkJYQwytK0wOj7r9dtz-v+YBW+1WbFJUZtu+xOLcN9S2NUr2HbXgG39f62QY-CWIfWO2p4UO5poqtGyjuuFlDh+G317QDSBlF0gY4y7XgHXc4KP1eCvlXil8YdpPiZYmtnwlz6L2qd8Je9-GFQEkJZniX-oMKdH7EYbXFIL6IkjWOW5kVx0KWXZFmUEJbkOcveEYm+sfiWupw58QNamXaiHduu1aOuMH7vUct0N0rbken7ue87NeCzQfmbmLvlZFDWujEzQFMDj0I6MGHDtXe5c+nvgjrlwo+iQTnVQjl0JfnCfqswjcObfW4kGL2EhRKTKGIkyrOU5X-neLoqNm9M-fcS+1AavAOJfsuwUw4b6-QqbCvrC99PUzS+fZjejevxvR-ZQZR9M4dJi0aGPwu1aNuI4g+z9iJCCSuKIknTwOBsjXzIN-5EZP+U1KBThu92p-07ip-k45Mf5dpl3m6p-yf1NtmgU-Y-+o9lB52-AM6IAuc=

[CableChannel]

@carloscadu: I sent you some updated snapshots via PM. Cheers