Are you beginning to see why it’s a terrific idea to work in a quiet studio with a good microphone? You can pile on corrections, tools, and filters, but they all distort the performance a little in addition to doing their job. Then you can pile on more tools to correct the distortions…etc.
ACX/Audible has a publication submission rejection called “Overprocessing.”
And yes, if you noise reduce twice, it will try to reduce twice, but the reduction distortions go up twice, too. Sometimes you can get good reduction by applying gentle reduction twice instead of hard reduction all in one shot.
And I am still thinking "Why do noise reduction?” in the Audacity sense of the menu command Effects, Noise reduction.
The only reason I can come up with is “Because some human says that they can hear noise”.
I am in the amateur business of recording audio books.
If I or someone else can hear a constant noise in the background, then I should turn off the heater-fan and record the track again. (This is an argument for recording one page at a time and then using macros to glue the pages into a chapter)
The key words here are surely “human” and “hear”. The bottom line (for me) must be, that if you can’t hear the room-noise or background-noise or hum or whatever, then there is no reason for me to consider applying noise-reduction.
As an amateur I am blessed now because I live alone in a small cottage on a quiet non-commercial street in a small town on the desolated end of a narrow peninsula jutting into the North Atlantic Ocean. Were I still living in the downtown core of North America’s busiest city for condominium construction, I would be singing a different tune.
To some degree there will always be electrically-generated self-noise, (not audible in the room).
On low-budget equipment this electrically-generated noise could be loud enough to require noise reduction and/or noise gate.
Thanks Trebor. I am horribly out of my depth in a way.
“(not audible in the room).” suggests to me that, as I sit here, my ears cannot detect a noise. I would say “this room is dead quiet”, which really means “my ears/brain do not report a signal”. That might be because my brain is so used to a signal that it blocks it out as “no change”, but we say “Chris cannot hear a noise, therefore there is no noise”.
Now if Trebor visited, Trebor’s ears might be better than Chris’s, and on entering this room, Trebor might pause and say “What’s that noise?”, the noise being new to Trebor’s ears/brain. In this case Trebor might advise me to apply noise reduction.
But if Trebor’s ears do not signal a noise, then I still say “No need to apply noise reduction.”
Finally, suppose that Chris and Trebor are both more than 75 years old, and both appreciate that their hearing might be failing.
Now they apply technology to analyze the recorded track, and the spectro-wossit displays noise signals.
At that time, Chris says 'Well I never!" and realizes that while his recorded tracks sound fine to him, some youngster might find that noise distracting.
Only if “this electrically-generated noise could be loud enough” to distract the human population should Chris be thinking about noise reduction.
Does that make sense?
I suspect that back in the days when overseas families mailed cassette tapes back and forth, no-one really cared about noise reduction at that level.
Electrically-generated noise from the (audio) electronics is inevitable*, and will always be present on a raw recording, even if it is not conspicuous on playback, (*it’s a physics thing).
Cheaper consumer audio equipment may generate a noise level of ~50dB, which is OK for Skype, but too loud for a professional-quality recording: ACX specify <60dB … https://blog.acx.com/2015/06/11/all-about-noise-floor/
Noise reduction is the last resort to bring the noise down to acceptable standards, as it can make the recording sound computery. If used subtly you can lower noise by ~6dB without it being obvious that noise-reduction processing has been applied.
@Trebor “Noise reduction is the last resort to bring the noise down to acceptable standards …”.
As each day passes my feeling increases that, because “Noise Removal” is on the menu at the Audacity 3.1.3’s café, everyone wants a heaping plate of it. Over at LibriVox I completed recording my first 13 tracks (6h45m recorded audio) in 17 days and have not had a comment on “Noise” from any proof-listener there. That makes me wonder whether people are trying to effect noise removal when they don’t need it. I have pointed out that I am graced with a quiet room for recording. I could not have recorded Audio at my previous place.
“Noise reduction is the last resort to bring the noise down to acceptable standards …”.is a valid statement, and to the point, it supposes that there is noise (in the background) and it is judged by humans to be at an unacceptable level. I am now of the mind that in over 50% of the cases, people are using noise removal in Audacity 3.1.3 because (a) it is there and (b) it is free.
I am now convinced that noise removal is NOT free; it comes at a measureable cost - that of distortion of the target signal.
@SecretCode “…even if I can’t hear something on a recording, it doesn’t mean that professionals with better ears won’t immediately detect it.”
Quite so! Ten years ago a free hearing test told me that I was at borderline level of 4,000cps in my right ear. Five years ago the same business told me nothing had changed.
Here I am recording audio books that people will listen to through ear-buds during their commute. I would bet that there is not a university within a thousand miles of me studying Noise Removal as a Post-Recording Technique in Audio books. There’s no grant money for it!
So, to the majority of us cheap people who want professional facilities in a superb yet free product, who are we trying to fool?
I am interested in “the noise removal problem” in the sense of working out “the conditions for effective use of the Audacity 3.1.3 noise-removal facility”; but as mentioned above, no proof listener has asked me to reduce noise in my recordings.
I am close to the point of thinking that the clamour for a better way of using Noise Removal is a red herring; a false trail.
Using Noise Reduction (if at all) is a balancing act between the amount of noise removed vs. the amount of damage done by Noise Reduction. It is very common for inexperienced users to overdo Noise Reduction and end up with a recording that has deathly silences and metallic jangly audio. For natural sounding audio it is generally better to use as little processing as you can get away with.
Thank you, Steve.
I spent part of yesterday and today superficially analyzing posts in the LibriVox forums. At LibriVox we amateurs record text from eBooks to audio. One can “Read” text to audio and/or one can “Proof-Listen” (PL) the audio. Always two people involved; I Read some text, and then a PL - an independent member - listens to it and asks for corrections to be made.
The MSWord2003 document at www.chrisgreaves.com/Downloads/WednesdayFebruary092022.doc is a summary of my findings with links to specific posts in LibriVox. (The Excel link is local to my laptop, and I have three typos of little consequence).
I searched LibriVox for the full string “apply noise reduction”; phpBB advanced search is broken and it returns partial matches. Even so only 0.114% (about 1 in 9,000) of the posts on the board talk about noise reduction. About one post every 26 days. And those posts are NOT PLs requesting that noise reduction be applied. AND I suspect that if I were able to track down such posts, a high proportion would be from PLs who have visited the Audacity cafe and seen “Noise reduction” on the menu.
My conclusion: I suspect that I, like many others, have been swept up in the belief that Noise Reduction will solve a problem that does not exist.
I have (happily) spent much time building a one-click macro that effects noise reduction, and (happily) learned a great deal about Audacity in the process, but my conclusion above, unless countered by a rational argument on LibriVox, will probably see me posting an awkward topic over there suggesting that members stop worrying about noise reduction unless and until one member, a PL, comes up with a specific track that exhibits noise whose effect can be shown to be improved by application of Audacity’s noise reduction command.
@ALL I am not suggesting that Audacity’s noise reduction command is useless; not at all. I am suggesting that Audacity’s noise reduction command is probably useless to LibriVox members who have been using it “because it is there”.
@ALL The Audacity phpBB site does not support the polling feature, but if it did, I would post a poll that asked “For those of you who listen to podcasts and audiobooks, how many of you listen through earbuds on your daily commute or your daily jog/cycle?”
phpBB Advanced Search is faulty and regularly mis-reports the number of genuine matches. That said a search for “de-essing” yielded “97 matches”. An Advanced Search reporting Topics Only returned 45 matches.
On a related issue I am contemplating turning my Selenium web-searching code loose on the LibriVox forum to determine serious requests from a Proof Listener to “apply noise reduction".
To my mind if I record a text and the Proof Listener passes it as “PL-OK”, then I should not start slapping extra coats of gloss paint on it. Only a specific request from a PL counts; and even then I am not sure that all PLs are “up on” what noise reduction can fix. (“Throughout the reading it sounds like traffic in the background”)
I search LibriVox for “apply noise reduction” and come up with only 227 matches out of 1,984,532 posts. Posts, not Topics/Threads/Projects, let alone Sections.
The difference in post numbers between highest and lowest of the 227 posts is 1,984,532 so the 227 posts represent about 0.0114%, or about one in nine-thousand posts.
The difference between post dates is sixteen years. About one post per month! And the post is not necessarily about “apply noise reduction” (phpBB search is broken).
As lax as I was, I still come up with only once-per-month.
Since the Selenium search procedure is automated, I could as easily search for requests from PL’s for de-essing. Or any other specific post-recording task.
I love the brain and harvest Richard Dawkin’s and Steven Pinker’s books on the brain.
After less than two months using Audacity (as the amateur voice recorder that I am) I am moving away from a position of awe that there are so many things I can do to my audio track, to a position of basically ignoring it until told to do something about it. I am “in the business of cranking out presentable audio books”.
In particular, using an Effect that I don’t know or understand seems akin to throwing a match into a steel drum and seeing if flames come out.
I suspect that the majority - perhaps 90% - of LibriVox members are like me - we want to record audiobooks but know nothing at all about audio quality.
Sound engineers we are not. Retired or occupied, yes (bus-drivers, lawyers, …) but sound engineers No.
De-essing is more like hammering-down any protruding floorboard nails.
Sibilance is a high-frequency thing,
so depending on your headphones/speakers/hearing you may not be able to evaluate it by ear.
Your January 2nd example does benefit from de-essing: you occasionally have a loud whistling ess …
Thank you Trebor.
When I listened to your treated sample I could hear that the essses were reduced, but was tickled that you (I thought) sounded just like me. I had forgotten that I had posted that sample!
I have known for years that I have much (or high?) sibilance
depending on your headphones/speakers/hearing you may not be able to evaluate it by ear
Relevant to my thesis that over at LibriVox we ought not to apply effects until told to do do, this means that a 75-year old PL who uses ear alone to PL, possibly will not detect the Essing (or other problem), whereas another PL with snazzy headphones and the hearing of a 20-year old may pounce with a pronouncement.
Thus on a single project (book) with multiple PLs, I might submit a track that passes PL, whereas had I submitted it the day before, it might not have passed.
I am excluding from “apply effects” LibriVox’s mandatory levels of volume and things like bit rate.
Me being me, once a PL pronounces that I have Essing, I ought to include de-essing in my application of Effects on all my tracks; and in this sense Trebor is a PL who has just pronounced Essing on me, so that is good to know.
I am at the point of saying that “we ought to apply Effects only to those matters that someone else has told us about”, and then that “In some cases it may be good to apply these specific effects automatically”
I am slowly getting there, Thank you, Trebor
Trebor, I am now officially out of my depth . I know what “Hz” is, but have no concept of what it means with respect to my recording of my voice.
I am still at the point of trying to make my audio recordings within a specified decibel level!
I will have to come back to this Hz business at a later date.
Possibly you have used a 16kHz sample-rate at some point during the recording/processing,
that would remove everything above 8kHz.
My January 2nd recording for LibriVox was to have them check that my hardware/software setup was a good profile.
Apparently it was.
Since then I have thought of two suggestions (for LibriVox)
(1) Besides the hardware/software settings, institute a second one-minute test for things like volume (amplification), pauses (between words, sentences, paragraphs etc), that is, those personal traits that are not picked up by examining a wave, but can be adjusted with experience.
LibriVox are very receptive and do not demand a flawless accent; with my mongrel tones from English prep school, Aussie goldfields, and Canada/USA, I am (gratefully) proof of that tolerance.
(2) Someone who records for two different audiobook publishers would have two different hardware/software setups and flip back and forth. Periodic hardware/software checks might be a good thing.
LibriVox does have a Checker.EXE program that inspects loudness, bit rate, all lower-case letters in filenames and so on. The author of Checker.exe has left the forum, so Checker.exe is an unsupported program right now.
This all leans me toward an attitude (on LibriVox) of “If it ain’t broke, don’t fix it”.
The sample-rate defines the maximum frequency that can be recorded, and is specified in Hz.
Your January example mp3 has a sample-rate is 44.1kHz, with bit-rate of 128kbps,
so is exactly as specified by librivox, however you’ve only used half of the audible frequency-range available: there’s nothing above 8kHz.
Somewhere in your recording/processing chain something has removed frequencies above 8kHz.
e.g. if your recording-device was set at a sample-rate of 16kHz, that would filter-out everything above 8kHz.
To avoid that in future recordings it’s just a matter of resetting the recording-device sample-rate to 44.1kHz.
If you are recording using a computer, Skype/Zoom type applications could be responsible for the 16kHz sample-rate.