Spoken Word: Tweaking for Maximum Quality

[Kevin.huska]

I first run the Hi-Pass Filter, setting the rolloff to 24db and the frequency to 165Hz.
Next the Low-Pass filter, setting the rolloff to 3db and the freq. to 3250. ...
I always run Noise Removal last.

That is how I would do it, although (depending on the original recording) I may prefer to set the lowpass filter a little higher.

As a point of information, the 3dB setting is actually 6dB per octave - they have been incorrectly labelled, but I believe that this has now been corrected in the current alpha version.

I leave the q value alone for both filters. I do not quite understand the function of this parameter.

Perhaps you would like to add support to my "feature request": http://audacityteam.org/forum/viewtopic.php?f=20&t=7584

Steve, I know this was a number of years ago. I just started using Audacity 1 week ago. What you and ogelquaeler talk about here has helped me.

I am using Audacity 2.0 on a Mac with OS X

Just wondering, when you say you would "set the Lowpass filter a little higher", do you mean that you would set the cutoff frequency higher or that you would set the Rolloff higher? If it is the Rolloff, how much higher would you set it? I used cutoff Frequency =3250, and Rolloff = 36 and it sounded ok... but wonder if my ears might be missing something. So just wanted to clarify what you meant by "Set the lowpass filter a little higher".
Thanks in advance for your help.

[steve]

Steve, I know this was a number of years ago.

Wow, it certainly was

Just wondering, when you say you would "set the Lowpass filter a little higher",

I mean set the Low-pass filter frequency a little higher.

3kHz is about the highest frequency on a traditional telephone land line, but you will probably be aware that sometimes it can be difficult to distinguish the differences between an "f" and an "s" over the telephone. The "main" frequency range in speech is around 300 Hz to 3000 Hz but some sounds ("s", "f", "t", "th") have frequencies that go quite a bit higher and these higher frequencies can be important for recognising these types of sound.

One study I read suggested that for maximum intelligibility the frequency bandwidth should go up to 7 kHz, though if there is a lot of hiss in the recording it may be better to limit the frequencies to less than that. Try experimenting with different frequency settings for the low-pass filter and see which works best. On moderately good recordings you should find that a 7 kHz corner frequency gives noticeably clearer sound but if the recording quality is poor it may be more intelligible set closer to 3.5 kHz.