Faint squeakiness in exported MP3 files

[Jim McC]

I have been using 1.2.6 for several years, but I am now trying 1.3.12 on my new PC that runs Windows 7. I use Audacity to edit and convert sermons to MP3 and post them on my church's website. When I export an MP3 file at a 32K bit rate, there is a faint, high pitched squeakiness in the MP3 file that did not occur in 1.2.6. If I go up to a 128K bit rate the problem goes away, but the resulting file is much too large for my purposes. Any suggestions about what I'm doing wrong? Thanks for providing Audacity. It's a great program.

[waxcylinder]

64k is generally regarded as the minimum quality for spken word recordings 0- 32-bit if you are in mono.

For music 128k is normally regarded as the minimum quality setting (64k for mono)

Otherwise you will hear digital artifacts as a result of the compression - those with really good ear will still hear artifacts at the levels above - note that Apple which used to sell on iTunes at 128 now sell at 256k

WC

[Jim McC]

WC
Thanks for your suggestion, but the problem was something else. I agree that a low MP3 export sample bit rate results in distortion, but I would describe a recorded voice as "muddy" due to the loss of the high end, not "squeaky." I would describe my "squeaky" probem as how an old cassette player sounded when you were in playback and slowly pushed the mechnaical fast forward button (yes, I'm really showing my age). For a moment before the tape head retracts, the tape starts to speed up and the output is "squeaky." My Audacity MP3 output has this kind of squeaky sound, but it is very faint on top of the intended voice signal.

However I have found a solution, more or less by trying every adjustment in Audacity, and that is to lower the project sample rate. I'm now at a project sample rate of 11025 with an MP3 export rate of 32K and the squeakiness is gone. True, the output is a little "muddy" due to the 32K MP3 rate but it is still very intelligible speech and the file size is acceptably small for posting on a website. Perhaps my post will be helpful to someone else with the same problem.

If one of you smart dudes could explain why my solution works, I'd find it very interesting.

Jim McC

[steve]

Your solution works because when you down-sample the track to 11025 Hz, that automatically limits the highest frequencies to about 5 kHz.
At 32 kbps, the current version of Lame will attempt to maintain a bandwidth up to about 8kHz, but due to the lack of "bits" it does so imperfectly. Data compression is always a trade-off between various aspects of "sound quality" vs bit rate (file size). These trade-off have been tweaked many times by the developers that produce Lame in an attempt to strike the best balance for "typical" audio.

A high bandwidth in the MP3 file will give a better perceived "brightness" to the sound, but too high a bandwidth will create peculiar squeaks as you have noticed - the exact balance to produce the best result is fine and depends on the type of audio in the recording. For music, it is probably better to have a slightly higher bandwidth than for speech, (a little bit of squeaking is possibly better than the music becoming too dull) and MP3 compression is optimised for music.

Because you have down-sampled before encoding, you have removed frequencies in the 5 kHz to 8 kHz range, and thus eliminated the squeaking.
You should find that if you used a "48dB per octave, 5000Hz" low-pass filter instead of down-sampling to 11025, the result will be very similar.
You could also try using a low-pass filter set to "48dB per octave, 6000Hz" - this may still have the desired effect of preventing the squeaking, but sound a little bit brighter.

An alternative to using MP3 is to use "Speex" http://www.speex.org/
Unfortunately Audacity does not itself support Speex, so to use it you would need to export as WAV, then use a stand-alone Speex encoder.
Another disadvantage of Speex is that a lot of people have never heard of it and would not know how to play a Speex encoded file.
The major advantage of Speex is that it is optimised for speech and can provide much better sound quality in much smaller file size than MP3 can.

[Jim McC]

Stevethefiddle: Thanks for your explanation. I tried your suggestion of using a higher project rate and applying a fast roll-off filter and I found the results to be similar as you say. But using the 11025 project rate seems just slightly better.

I'd appreciate one more explanation from one of you smart guys. My original audio source is mono (a PA system) and recorded on both stereo tracks of a CD. I bring the CD home and use Windows Media Player to rip the CD and make a wav file to use as the Audacity input. I would have expected that converting the Audacity project from stereo to mono would have no effect since I am converting two identical signals into one signal, so there is no change in the amount of information. I tried this anyway and was surprised to find that the exported MP3 file (still using 32K bit rate) had better quality. It sounds like it has more high frequency content, eliminating most of the "muddy" sound that resulted from the low bit rate. The resulting MP3 file size is the same. Thoughts? This technique might be useful to others who want the best sound quality in a very small file to post on the web if mono is OK for their purposes.

Jim McC

[steve]

I would have expected that converting the Audacity project from stereo to mono would have no effect since I am converting two identical signals into one signal, so there is no change in the amount of information. I tried this anyway and was surprised to find that the exported MP3 file (still using 32K bit rate) had better quality.

Whether you encode a single channel (mono) track at 32kbps (kilo bits per second) or a 2 channel (stereo) track at 32kbps, the MP3 file will still have 32 thousands binary digits (1's or 0's) for each second. In the case of the stereo track, the 32 thousand bits are shared out between the two channels, but for a mono track all 32 thousand can be used for the one audio channel. This means that the mono track will be roughly the same sound quality as a stereo file encoded at 64kbps.
(there is a certain amount of "overhead", so it does not work out quite as exactly as this, but it's close).