Active TopicsActive Topics  Display List of Forum MembersMemberlist  CalendarCalendar  Search The ForumSearch  HelpHelp
  RegisterRegister  LoginLogin
General Discussion
 FSI Language Courses Forum : Learning Languages : General Discussion
Message Icon Topic: digitalization Post Reply Post New Topic
<< Prev Page  of 2
Author Message
gdfellows
Administrator
Administrator

Site Creator

Joined: 04 March 2006
Location: United States
Online Status: Offline
Posts: 237
Quote gdfellows Replybullet Posted: 04 September 2006 at 9:31pm
Originally posted by r438

Also, anyone know what dpi setting the other pdfs were scanned at? 
 
I've scanned all PDFs here at 300 dpi.  I do use the "reduce file size" option in Adobe Acrobat.


Edited by gdfellows - 04 September 2006 at 9:32pm
IP IP Logged
DemiPuppet
Administrator
Administrator


Joined: 27 May 2006
Location: United States
Online Status: Offline
Posts: 163
Quote DemiPuppet Replybullet Posted: 05 September 2006 at 9:55pm
I think a sharp lowpass filter at 8Khz is best, otherwise some of the sibilants will get lost.

http://en.wikipedia.org/wiki/Sibilant

If the recording is in stereo, make sure it's converted to mono.  This will reduce the hiss by a few dB.

If the recording is really miserable, you could try using an expander.  For example, some of the Turkish tapes were recorded at a really low audio level. In order to pull the sound out of the hiss, I would amplify only the audio with an amplitude greater than the hiss level.
  1. Using Audacity, sweep out a couple of second of non-speaking hiss on the timeline.
  2. Select "Analyze|Plot Spectrum
  3. Note the audio amplitude in dB's around 4000 Hz
  4. Determine how much the audio can be amplified without clipping.
  5. Amplify the audio that is louder than the value found in step 3.
I use the "normalize" commandline program and the "-n" option for step 4 and the "sox" command line program and the "compand" option for step 5.

http://normalize.nongnu.org/
http://sox.sourceforge.net/

Since the Turkish 2, Vietnamese, and Swedish tapes weren't too bad sounding, I didn't perform this step on them.   Probably should have.
IP IP Logged
Sir Nigel
Contributor
Contributor


Joined: 16 March 2006
Online Status: Offline
Posts: 35
Quote Sir Nigel Replybullet Posted: 06 September 2006 at 9:09am
DemiPuppet, how would converting to mono alone reduce hiss? Is it by perception only?

Obviously stereo is pointless for all FSI courses, but I would imagine the "rubbish in, rubbish out" rule would apply unless you're using a noise reduction plug-in.

Originally posted by gdfellows

I've scanned all PDFs here at 300 dpi.  I do use the "reduce file size" option in Adobe Acrobat.


In addition I've saved my PDFs using JPEG 2000 and the file size is tremendously lower than regular PDFs.
Insurance Excuse:
The accident was due to the road bending.
IP IP Logged
DemiPuppet
Administrator
Administrator


Joined: 27 May 2006
Location: United States
Online Status: Offline
Posts: 163
Quote DemiPuppet Replybullet Posted: 06 September 2006 at 10:15pm
The assumption is that the hiss is in each channel is totally random while the desired signal is nearly identical.  So when the left and right channels are added together, the total "good" signal is doubled (+6dB).  But since the noise is random, sometimes it adds together and sometimes it subtracts. Totally random noise will add as a "Root Mean Sum" and only increase by 3dB.

When converting to mono, the two stereo channel signals are summed together and then divided by 2 (in other words reduced by 6dB). So now the mono "good" signal is back to it's original level, but the hiss is reduced by 3dB. The hiss amplitude will be reduced to ~0.707 of its original value.

As a bonus, going to mono means that the precious bits used by the MP3 encoder can be dedicated to just one channel rather than being wasted trying to also encode the differences between the channels. I suspect the original recodings were all in mono anyway.

One additional way to maximize the use of the MP3 bits is to filter out all the high frequencies (>8Khz) and then downsample to 22050 Hz.  According to the Nyquist-Shannon's sampling theorem, the sampling rate needs to be only twice the highest signal frequency in order to perfectly reproduce the sound.  By reducing the number of samples, less MP3 bits are needed to produce the same quality of signal. Of course you could keep the same MP3 bit rate and thus improve the quality.  I know it seems odd that reducing the sampling rate would actually improve the final signal quality, but it's true.

I use the SOX command line program to do the resampling after filtering out the high frequencies.

IP IP Logged
<< Prev Page  of 2
Post Reply Post New Topic
Printable version Printable version

Forum Jump
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot delete your posts in this forum
You cannot edit your posts in this forum
You cannot create polls in this forum
You cannot vote in polls in this forum



This page was generated in 0.141 seconds.