I’m using Audacity to record the nuisance barking of a dog. I set up the programme using timer record to start up early in the morning and allow it to run for several hours so I get quite a long recording. I can then later search through this recording looking for the barking episodes but it can take a very long time to do.
I’ve been struggling for quite a long time now to find some way of automating this procedure using Sound Finder but I’m not really getting anywhere with it can anyone suggest how I can do this please.
When this dog barks each barking event, which may last several seconds or minutes, consists of several individual barks, say 11 in 5 seconds, and I don’t want to identify each single bark just the group.
I’ve attached a very brief .wav file as an illustration of just one event and any help or tips anyone can give will be very much appreciated.
My setup
Audacity version 2.0.5 obtained using the .exe installer
Operating System. Windows XP Build 2600.xpsp_sp3_gdr.12054-1619: Service Pack 3
I cannot set the parameters in Sound Finder so that it selects and labels just the barking groups. I sometimes get literally hundreds of selections far too many to Tab through and often times I get selections made where there is no barking at all.
For some reason I cannot get the sample .wav file to upload Steve I’ve tried adding it to my reply here using both Google Chrome and Internet Explorer. I’m fairly used to the upload procedure and can’t understand what’s going wrong. I type my reply then select the file for uploading and it shows that it is uploading but as soon as it reaches 100% I’m switched immediately to a new blank reply page with only the Subject line filled in, nothing else. Tried all this with my Firewalls switched on and off same problem every time. If it is uploading I’ve no idea where it’s going to!
Well it hasn’t uploaded. I’d guess that it is some other security/privacy setting that is preventing the page from reloading properly. For the sake of progressing your enquiry I’d suggest that we bypass this for now. Try uploading your file somewhere else - I usually use sendspace.com as it allows up to 300 MB files without the need to register or log-in. The downside of sendspace is that they sometimes have sneaky adverts/installers, so be careful what you click on, and/or use an “ad-blocker”.
If you could send a couple of samples, one with the noise that you want to detect, and one with background sound that you want to ignore, I’ll try and help you find good settings (if it is possible for silence finder).
The difficulty in detecting the dog barks is that it is much quieter than much of the other background noise (though I presume it is more irritating for you than the other background noise). This fact will considerably weaken your case if you are wanting to take action against the dog owner, but that’s not your question, so on to the task in hand.
You can filter out most of the noise that is not the dog bark by pasting this code into the “Nyquist Prompt” effect (Effect menu) and applying to the audio track:
(highpass8 (lowpass8 s 800) 600)
Note that this will not be very quick, so you may want to try it out on a short selection first.
The result will be very low level, so next you will need to boost it using the “Normalize” effect. Set “Normalize maximum amplitude” to -1 dB.
Then use “Sound Finder” with the following settings: Treat audio below this level as silence [ -dB]: 15 Minimum duration of silence between sounds [seconds]: 5.0 Label starting point [seconds before sound starts]: 0.0 Label ending point [seconds after sound ends]: 0.0 Add a label at the end of the track? [No=0, Yes=1]: 0
Again this will be quite slow.
To verify that the labels are dog barks and not other sounds, click in a label to select it and press ENTER to move out of text edit, then press spacebar to play.
Tab, then Enter to move to the next label.
Note also that the filtered and amplified recording is not acceptable as “evidence” in any legal sense because the evidence has been tampered with.
Many many thanks for your reply it works wonderfully well. I carried out an “analysis” of a 10 hour recording I made the other day and the procedure took just over one hour which would have taken me two or three hours to do manually.
The result was that all the barking episodes were selected and labelled and of course there were a number, but not a lot, of other noises that were picked up but that presents no problem at all. I think I grasp the various parameters in this process you’ve set out and I can twiddle them at leisure to see if it improves the result.
Incidentally this process of evidence gathering I’m going through has nothing at all to do with any sort of legal situation and I’d be happy to explain if there’s some way I can private message you, you’ll understand I wouldn’t want to say more in a public forum. Please rest assured there are no dark underhand or devious motives at work here.
One again many thanks for your help in this it really is most appreciated.
Many thanks for the link. Couple of things if I may.
In the instructions you gave me I guess the line ““Normalize” effect. Set “Normalize maximum amplitude” to -1 dB” amplifies the recording so that the Sound Finder can pick out the barking events better, could you give me an instruction that would bring them back to something like the actual recorded level after I have extracted them and exported them as a .wav file?
Something that’s a little bit more of a problem. In yesterday’s 13 hour recording the procedure failed to pick out 5 barking events although they looked almost identical in amplitude and length to the many others it did find. If I make up a small file containing some of the unidentified and identified ones could I be a nuisance and ask you to have a look at it for me please?
“File menu > Import >Audio”
Select the original file. It will be imported below your modified version.
If you want to hear only the original file, click the “Solo” button on that track, (or click the “Mute” button on the modified track).
The Normalize effect brings the maximum amplitude up to the specified level, which in our recipe is -1 dB.
Even after filtering, the loudest sound (highest peak level) in the Sunday file is your voice (I presume that is your voice).
Therefore, when we Normalize, it will bring your (filtered) voice level up to -1 dB rather than the dog’s voice level, so now Sound finder will mark all the places where you are talking, but miss out the dog’s singing.
Steve I made all the extracts and exported them first of all. Then I created a new file and imported them and only at that point did I record the voice identifying which extract was what. I did not try to run the procedure after adding the voice description.
Do you want me to prepare another file with only the extracts and no voice?
Thanks for your mail I think I might have confused things with the file I uploaded, sorry about that. I’ve got to be honest and say I’m not all that clear on what your file demonstrates except that your use of the recipe found the sounds whereas on my application of it to the original it didn’t for some unknown reason.
Anyway since looking at your file I’ve been experimenting with the original .aup file and I find that by progressively increasing the decibel limit the recipe finds more and more of the missed barkings. I find that by setting the level at 16.9, which maybe executes at 17, the recipe found all the missed events.
It did of course also pick up quite a lot of other extraneous sounds but I can live with that, I’d rather have too many of those than miss any barking events.
I’m sorry to be a damned nuisance but could I ask you to have a look at this little file and see if you have any idea at all why Sound Finder can’t pick the sounds up please? Just two events from today separated with a few seconds silence.
