Forums | Search | Archives

All Forums  Cool Edit  Recovery of speech buried in tape noise

Author	Topic
ForensicMan Posts: 2	Posted - Mon Aug 04, 2003 8:33 am  I have a extremely badly recorded audio cassette and need to extract the speech audio, an interview, as clearly as possible for transcription. I have captured the cassette audio on the PC as a WAV file. I have played with the noise reduction controls of CEP2.1 but could some kind and experienced person give me some tips for extreme recovery of low level speech buried in tape noise. Tim
Rhino Posts: 14	Posted - Mon Aug 04, 2003 10:31 am  I do alot of noise reduction to remove a slight hiss or hum from a track. I try to fing a part with only tne unwanted noise and select it. Then go to noise reduction and click on Get Profile Form Selection. Then I use this profile on the complete wav. in almost all cases it works well without altering the sound alot. Hope it helps...... _________________ Tri-R Studios
Graeme Member Location: Spain Posts: 4663	Posted - Mon Aug 04, 2003 10:53 am  "Altering the sound a lot" is not really an issue for this sort of thing - providing it with intelligibility is. My experience is that CE is not the best tool for this sort of forensic work. There are better tools around, such as Sound Cleaner Pro. However, at about three times the cost of CEP, you will have to dig a bit deeper into your pocket. _________________ Graeme Don't forget to join the new CEP forum at audiomastersforum
ForensicMan Posts: 2	Posted - Mon Aug 04, 2003 10:53 am  Thanks. The speech I am trying to recover really is so low level that it is barely audible above the noise floor. The recordist put an unsuitable microphone much too far away from the interviewees. My only objective in this task is to be able to decipher the words spoken more clearly.
MusicConductor Location: USA Posts: 1524	Posted - Mon Aug 04, 2003 11:03 pm  ForensicMan, just to reinforce and clarify Graeme: Cool Edit does a really great job of removing noise, and many of us use it for restoration or correction. However, there are few occasions I can imagine it would improve intelligibility, but rather works parallel to it, for lack of better terminology. It's always been a matter of "garbage in, garbage out," which is why VooDooRadio's First Law of Noise Reduction is true: Noise Reduction works best on signals with... very little noise. Ummm, SWMBO just summoned me. Bye!
VoodooRadio Location: USA Posts: 3971	Posted - Tue Aug 05, 2003 8:22 am  Quote: MusicConductor Posted --------------------------------------------------------------------- Ummm, SWMBO just summoned me. Bye! Careful!! _________________ I said Good Day! Voodoo
bonnder Posts: 215	Posted - Tue Aug 05, 2003 9:20 am  ForensicMan wrote: The speech I am trying to recover really is so low level that it is barely audible above the noise floor. My only objective in this task is to be able to decipher the words spoken more clearly. Would the Frequency Band Splitter help at all here? Wouldn't the spoken word be more pronounced in certain frequency ranges than others? Load your file into Multitrack view. Highlight your file. Click on "Effects / Frequency Band Splitter". Click the radio button on all eight frequency ranges to activate them all. Use the defaults or put in your own. (Might help to review the "Frequency Band Splitter" entry in the Help file also.) You might discover that your words rise more above the noise floor in a certain band or bands. Work with that frequency band(s) and see if noise reduction techniques work a bit better there. I know somebody out there can fine-tune this process better than I have laid it out here. MC?? Are you back yet?
MusicConductor Location: USA Posts: 1524	Posted - Tue Aug 05, 2003 12:00 pm  MusicConductor wrote: Ummm, SWMBO just summoned me. Bye! You know, things aren't always what they seem! (11PM is the time of little baby Sara's last feeding for the evening...) Now get this: my wife referred to herself as S.W.M.B.O.!!!!!!! ;) One thing I like about Bonnder's suggestion is that creativity is a great way of inventing solutions to unique audio problems. But I don't think the Frequency Band Splitter would reveal anything that the Graphic Equalizer wouldn't. The advantage of the latter is that it can be used in Edit View with a preview in real time, allowing the results to be assessed immediately. But Bonnder may have suggested one valuable thing: reduce from the audio frequency ranges that contain only noise. I'd recommend using Spectral View to look at several second's worth of sound at a time to get a general idea how much high end can be reduced without deleting some of the voice. This may still not increase intelligibility if the overall sound is perceived to be dull and muffled -- even if it is only noise that makes it not seem muffled!
bonnder Posts: 215	Posted - Tue Aug 05, 2003 12:31 pm  I'm focusing on the fact that ForensicMan wants to increase the intelligiblity of the spoken word so that it can be transcribed. I'm suggesting that perhaps there is a frequency range wherein the intelligibility is rather more and the noise is rather less - or at least maybe the noise can be reduced rather more successfully. Someone recently posted something about the spoken word having more energy in a certain frequency range (I can't find the post tho) and that is what I am building on. (A range which would differ, depending on whether male, female, child, etc.) In ForensicMan's situation, I'd be less concerned about cutting out highs or lows of the vocal range and more concerned about finding a frequency range (if there is one) where the intelligibility of the vocals rises a bit more above the noise. Frequency band splitter or graphic equalizer, I'm not pushing one over the other. Whatever works best for ForensicMan. And MC, SWMBO appreciates your edit.
MusicConductor Location: USA Posts: 1524	Posted - Tue Aug 05, 2003 2:04 pm
jonrose Location: USA Posts: 2901	Posted - Tue Aug 05, 2003 2:56 pm  ...and apparently, my deletion... Heh! _________________
DeluXMan Location: Canada Posts: 330	Posted - Tue Aug 05, 2003 3:31 pm  Which component, noise or speaker, is louder and by what ratio? Does any of the speaking show as tiny peaks poking out of the noise? Or is the level much lower than the noise? The frequency band splitter followed by noise reduction is a great idea. Success will depend on whether this raises the speakers' level enough on each band to use the noise reduction. If the noise is greater than the speaking, compression may work, or just boost the low level signals using the CEP dynamics, followed by noise reduction... OK i tried some things - i'm not sure if basic noise reduction helped at all, but i found that a multi-band expander worked a little bit, on a sample i made with pink noise plus enough monologue to recognise only the occational word. It brought out a few more words but only made a small change overall.
	Topic

Page: