Jens, I use the Quote function, i.e. the cartoon like speech bubble.
If there is a better tool I would be happy to use it. What you see in my posts, and always have done, is the simple quote
As you can see in the post earlier, I manually added the sources for clarity.
I don't see a problem here.
Different browsers, computers, Operating systems, make people see different things. Kinda ironic I guess ;-)
I sincerely hope that you have gained a full understanding of the fundamental difference between an Audio File of a Recording of a Sweep played loudly in a Room, as opposed to an Impulse Response. I can send you examples of both if you wish. The former is typically close to FS. The latter is often way lower in level, but not always.
EDIT Jim, good stuff, as you can see when normalisation is used the resolution of the Impulse Response is preserved through the Export/Import process.
On a quick glance it looks like your noise floor is 50dB below the initial spike. Most prosumer situations I see have background levels of 40dB or so. 90dB sweep and there's your 50dB range.
I prioritise driving the room hard. Your speakers could probably play louder without distorting. Wear closed cans or earplugs.
Jens, I use the Quote function, i.e. the cartoon like speech bubble.
If there is a better tool I would be happy to use it. What you see in my posts, and always have done, is the simple quote
As you can see in the post earlier, I manually added the sources for clarity.
I don't see a problem here.
Different browsers, computers, Operating systems, make people see different things. Kinda ironic I guess ;-)
I sincerely hope that you have gained a full understanding of the fundamental difference between an Audio File of a Recording of a Sweep played loudly in a Room, as opposed to an Impulse Response. I can send you examples of both if you wish. The former is typically close to FS. The latter is often way lower in level, but not always.
EDIT Jim, good stuff, as you can see when normalisation is used the resolution of the Impulse Response is preserved through the Export/Import process.
On a quick glance it looks like your noise floor is 50dB below the initial spike. Most prosumer situations I see have background levels of 40dB or so. 90dB sweep and there's your 50dB range.
I prioritise driving the room hard. Your speakers could probably play louder without distorting. Wear closed cans or earplugs.
DD
I suggest that you read the thread again or you´ll miss the point. I´m trying to understand why many users IRs are so noisy/truncated. A poor gain structure in conjunction with low playback level (speakers) is probably the cause and if bad gain structure; what can be done to improve matters.
I´m trying to understand why many users IRs are so noisy/truncated. A poor gain structure in conjunction with low playback level (speakers) is probably the cause and if bad gain structure; what can be done to improve matters.
I work with many clients remotely. I get them to measure using REW or FM. In general their levels are fine and the process works just fine. My own tests here show the same. Kissing FS on the Input Level meters or Interface meters, delivers ETC Magnitude peaks of only -20dB quite often. Chris has explained that one should not expect IR and Room Recording to correlate. ARTA is doing this differently.
EXPORTED Wavs are noisy and truncated because they have not been normalised.
As you can see from Jim's most recent Normalised, Exported, Imported, graph.
All is OK in the world.
Here are two more. I went into control panel in windows and turned the mic levels way up. I also have provided a normalized file for comparison.
So, as expected; simply normalizing does not improve the noise floor:
EDIT:
Quote:
Originally Posted by DanDan
EXPORTED Wavs are noisy and truncated because they have not been normalised.
As you can see from Jim's most recent Normalised, Exported, Imported, graph.
All is OK in the world.
DD
But there are some minor differences which is a bit interesting assuming the exact same measurement was exported.
When you say “I went into control panel in windows and turned the mic levels way up”, what exactly do you mean? There should not be any digital attenuation in a recording chain. What input peak level did your application indicate?
It´s the peak dBFS level of the recorded sweep and the S/N ratio of this recording that will determine the quality of the resulting IR file. If inadequate gain structure and/or low playback level (SPL in room); the IR will be noisy and possibly truncated if recorded/exported in 16 bits.
It´s the peak dBFS level of the recorded sweep and the S/N ratio of this recording that will determine the quality of the resulting IR file. If inadequate gain structure and/or low playback level (SPL in room); the IR will be noisy and possibly truncated if recorded/exported in 16 bits.
I agree entirely, but the statement is incomplete.
With 24 Bit recording, 144dB range, it would be extremely difficult to make an inept recording of a typical room which would struggle to have a 60dB range e.g. 95dB Sweep and a 35dB Ambience.
Even with 16 Bit recording, and hitting only -20dBFS on the input meter, one still has a recording with a 76dB range.
FM and REW both reject measurements taken with too low a recording level.
I kinda think John and Chris know what they are doing.
EDIT Jim, that recording normalised or not is very very noisy. Your ETC imported into REW or FM looks OK, the Waterfalls are appalling. Audibly it sounds like a combination of physical machine noise plus preamp/mic hiss. Something is very wrong here. I suggest you try some measurements with REW as we know this works well.
Do you have an SLM to measure how loud you are driving the room?
EDIT The introduction and in thread testing of a fairly unknown soft/hard ware package is an extra layer of confusion here. I posted here initially because FM was being IMO unfairly singled out as having a potential problem. This seems to persist.
J-
Quote:
It´s also perfectly possible that the only general “issue” is the 16 bit export without normalization (but then the application should be clear about the possible issues when doing so since most of the exported IRs from Fuzz I´ve found is a 16 bit un-normalized export and usually around -20 dBFS, that is causing problems).
As explained in detail REW and FM can identically Export IR Files converted to 16 to 32 Bit Wavs.
Both are equally capable of generating an unwanted problem if the designers recommendations are not followed.
Quote:
The most common mistake when moving impulse responses between applications is exporting the data without normalising it and compounding that by using a 16-bit format for the export. That throws away a lot of resolution, which will be apparent when looking at the imported IR and plots derived from it, such as ETC and waterfalls.
__________________
REW Author
DD
Last edited by DanDan; 12th December 2012 at 05:18 PM..
Reason: 4 Jim.
EDIT The introduction and in thread testing of a fairly unknown soft/hard ware package is an extra layer of confusion here. I posted here initially because FM was being IMO unfairly singled out as having a potential problem. This seems to persist.
So you´re accusing me of “single out” FM and at the same time don´t want to discuss other apps?
As the thread starter, I´m getting just a tad annoyed. I know you like FuzzMeasure (as can be seen on the FuzzMeasure home page), but this shouldn’t keep you from wanting to investigate a possible issue. Heck; even Chris (Fuzz) has stated that there might be something not right so what’s going on DanDan?
Quote:
Originally Posted by liscio
It's possible that the synchronous averaging is to blame for the lower level here because of the nature of my implementation, but it's hard to say here. I'd say to avoid it in favour of longer sweep durations if you're having trouble with SNR.
Why not let Chris have a look at it and report back once he knows something? And what´s wrong with testing FM and other apps in order to find out how they operate and how to minimize potential problems? If nothing is done, you would still find a lot (if not most) users ending up with IRs like these: Before posting your measurement results
When you say “I went into control panel in windows and turned the mic levels way up”, what exactly do you mean? There should not be any digital attenuation in a recording chain. What input peak level did your application indicate?
Control Panel / Sound / Recording / Microphone / Levels
Perhaps describing the Omnimic setup will help in troubleshooting here.
The Mic is USB, it plugs directly into my computer USB port. No mic preamp. No feedback loop capability. The Mic shows up in Windows in the path described in the above post.
I cant do today probably, but I will do another with all the other Windows mixer levels set to zero. DanDans description of "machine noise" sounds like maybe there are some windows artifacts getting into the IR recording.
Jens, both FM and REW (as stated by it's author) are equally capable of Exporting low res 16 Bit non normalised translations of the internal hi res IR.
We all see this issue and personally, I thank you for bringing it to notice.
However, each time you mentioned the problem, you associated it ONLY with FM. Similarly you presented a raft of Non Normalised 16 Bit Exported files from FM ONLY. I just think this is unbalanced and hardly fair to a professional commercial product or to it's designer.
EDIT. Of course I like FM, it's excellent, and I don't get it free. I have no hesitation in recommending products which I like or standing up for a friend.
Conversely, over time, I have brought issues to the attention of both designers.
There is absolutely no evidence which shows FM better, worse, or different to REW in terms of Recording level setting or the ease of Low Res Export.
It behoves us all to recommend Normalising prior to Exports. EDIT (and 24 or 32 Bit)
I do hope Chris and John may make that the default behaviour in future.
The occasional bug Chris mentions, where multiple sweeps may not add correctly, is a known issue. It may well be a downside to not using Loopback Correction.
I have no idea if REW will misbehave similarly without LC.
I have not experienced the bug in either app.
It might make more sense to lose the multiple sweeps in favour of longer ones as many people don't use LC unless doing critical work.
In reality it takes no longer to do multiple sweeps than a longer single.
EDIT Obviously IMO there is nothing wrong with testing. I have done and shown repeated tests here, which show no level setting problem or even a difference between them.
As we can see here FS in REW is equal to 94dB in FM and both tally exactly to incoming audio level.
Jens, both FM and REW (as stated by it's author) are equally capable of Exporting low res 16 Bit non normalised translations of the internal hi res IR.
We all see this issue and personally, I thank you for bringing it to notice.
However, each time you mentioned the problem, you associated it ONLY with FM. Similarly you presented a raft of Non Normalised 16 Bit Exported files from FM ONLY. I just think this is unbalanced and hardly fair to a professional commercial product or to it's designer. There is absolutely no evidence which shows FM better, worse, or different to REW in terms of Recording level setting or the ease of Low Res Export.
It behoves us all to recommend Normalising prior to Exports.
I do hope Chris and John may make that the default behaviour in future.
The occasional bug Chris mentions, where multiple sweeps may not add correctly, is a known issue. It may well be a downside to not using Loopback Correction.
I have no idea if REW will misbehave similarly without LC.
I have not experienced the bug in either app.
It might make more sense to lose the multiple sweeps in favour of longer ones as many people don't use LC unless doing critical work.
In reality it takes no longer to do multiple sweeps than a longer single.
EDIT Obviously IMO there is nothing wrong with testing. I have done and shown repeated tests here, which show no level setting problem or even a difference between them. Attachment 320978
DD
You still don´t listen to what I say so what should I say ... ? Nothing I guess is the best reply.
Perhaps describing the Omnimic setup will help in troubleshooting here.
The Mic is USB, it plugs directly into my computer USB port. No mic preamp. No feedback loop capability. The Mic shows up in Windows in the path described in the above post.
Ok, the gain of the built in preamp (in the mic) is controlled by the windows mixer? How do you check your recording level?
Here is a 16bit version just to see if looks different.
----------
edit:I just listened to the tracks in media player. I can hear what DanDan was calling machine noise.
There is a host of reasons I can think of that may not be the Omnimic or its software responsible here.
1) Cheap sound card. Using built in one chip device that came with mainboard
2) I am using 2 maximum length USB cables (10ft) with a USB hub between them before it gets to the PC
3) Windows Vista
In any case, the data I am getting seems to look good except the waterfalls. My own waterfalls show all this noise also. The CD tracks themselves sound clean to my ear, so its something between the mic and the recording. I may do some tests using one short USB cable at some future point to see if lengthy cables or the Hub are the culprits. But dont feel like fooling with that right now.
But I do thank you here Jens. Now I know something I didnt know before.
----------------
edit 2: I tried one short USB, no better. I tried at higher recording windows level, made it worse.
Still noisy, and again; normalized or not, in 32 bits or 16 bits, is not the issue here. You need to figure out how to achieve a better gain structure and/or play the sweep louder in order to increase the S/N ratio.
So, the mic “preamp” is now at maximum gain? How loud did you play your speakers? Was it close to painful without ear protection in the high midrange part of the sweep?
What peak recording level (in dBFS) did the software indicate when you made the recording?
Your sweep is still short. Use a longer sweep (6-12 seconds, log) and possibly also multiple takes (averaging).
Still noisy, and again; normalized or not, in 32 bits or 16 bits, is not the issue here. You need to figure out how to achieve a better gain structure and/or play the sweep louder in order to lower the S/N ratio.
I agree. This is a dead end until this part of the issue is resolved.
Quote:
Originally Posted by Jens Eklund
So, the mic “preamp” is now at maximum gain? How loud did you play your speakers? Was it close to painful without ear protection in the high midrange part of the sweep?
The mic "level" keeps resetting to 33 on a scale of 100, so I am not sure those graphs are any different than the others. I did get one set at 100 with no sweep at all and compared it to one at 33 with no sweep, and the backround noise was even worse at 100.
Quote:
Originally Posted by Jens Eklund
What peak recording level (in dBFS) did the software indicate when you made the recording?
85db, or so the FR graph says. My first IR wav was at 70db (for comparison).
Quote:
Originally Posted by Jens Eklund
Your sweep is still short. Use a longer sweep (6-12 seconds, log) and possibly also multiple takes (averaging).
A longer sweep wont work because the only other one I have takes longer than the fixed 620ms window. I am posing this problem at AVS where the software developer posts.
What the FR graph says is not relevant. I´m talking about the rec level meter you see when you record the sweep(s). How do you set your input levels?
EDIT: I’m going to have a look the OmniMic manual ... later.
Like I said, I have no record level meters. Every operation the software does except this one (Save IR) operates in real time and no matter how loud the volume level in the room, the graphs paint up and work (probably more accurately at louder volumes though).
Like I said, I have no record level meters. Every operation the software does except this one (Save IR) operates in real time and no matter how loud the volume level in the room, the graphs paint up and work (probably more accurately at louder volumes though).
Hmm ... that sounds a tad weird. I really need to have a look in the manual ...
Here is what the noise floor reads as with no test signal
Is the mic calibrated in terms of SPL at a fixed gain setting? This shows the noise floor of the entire chain, including your room’s ambient noise floor (assuming calibrated mic).
Is the mic calibrated in terms of SPL at a fixed gain setting? This shows the noise floor of the entire chain, including your room’s ambient noise floor (assuming calibrated mic).
New Reply 
Facebook
Twitter
LinkedIn
