From this webpage:
Get rid of the audio below 300-500 Hz*, and then use audio processing (compression) to raise the average power.
*This is accomplished by:
- Choosing a suitable microphone, and
- Equalizing the audio (if possible)
Audio which is primarily the mid-range portion of the human voice
(i.e. 300 to 2400 Hz.) carries better over long distances. (It is also
frequency-efficient on crowded bands). The downside is that this type
of narrow audio is not as enjoyable to listen to in a ragchew where
signals are S9+. The wider audio is more pleasant to listen to for
extended periods, but is not good for contesting, DX chasing, or use
on crowded bands.
It is therefore good practice to eliminate the low frequencies below
about 300 Hz, because intelligible speech does not require the
transmission of frequencies lower than 300 Hz. To do so adds
practically nothing to intelligibility. Elimination of the frequencies
below 200 or 300 Hz removes a large percentage of the high energy
speech components that do not contribute to intelligibility. (see Fig.
2-9 below). Such elimination permits the transmitter to concentrate
its efforts on only the essential portions of speech power. In
practice, this means something like a 3 to 6 dB improvement in system
effectiveness, equivalent to doubling or quadrupling its output power
even before any speech processing.
To increase the average power of the voice signal without increasing
the peak power, … by emphasizing the low-power, high-frequency
components of the speech signal, and attenuating the high-power,
low-frequency components of the speech signal.
Speech processing (compression) is most useful when we do this!
Figure 2-9 shows power-vs.-frequency distribution in the average human
voice over a range of approximately 200 Hz to 3000 Hz. This curve
indicates that the greatest concentration of speech power is at low
frequencies. Fortunately, it is the low-frequency components of speech
which contribute the least to intelligibility since these frequencies
generally occur in the vowel sounds. As a result, the low frequencies
may be attenuated without undue loss of speech intelligibility. The
low-power, high-frequency components present in a voice signal can be
pre-emphasized to provide some increase in the average power level of
the signal. Since it is the high-frequency components which
predominate in the consonant sounds, some emphasis of the high
frequencies improves intelligibility. In other words, it's far more
important for us to hear the S's, C's and high-frequency sounds like
those. If we don't, we may not understand what the guy at the other
end is saying!
Wattmeter needle swing:
Our power output meter and/or plate current
meter is NOT a reliable indicator of how well the other station hears
us! Fig. 2.9 below clearly shows why.
- There is more power in lower-frequency (bass) frequencies, so we observe more meter swing
- There is less power in higher-frequency (treble) frequencies, so we observe less meter swing (unless compressed or processed).
This is only a portion of the article. Much more information --with references-- can be found there.