Using the filters for podcasts and spoken word
High-pass and low-pass filters can be used together to concentrate the encoded frequency range on speech. This can reduce unwanted rumble or hiss and, at lower bitrates, avoid spending MP3 data on frequencies that contribute little to the programme.
For natural podcast or audiobook speech: a high-pass around 80–100 Hz is a useful starting point for reducing low-frequency rumble and plosives while retaining the body of most voices. Leave the low-pass on 0 (Auto) unless you have a reason to narrow the upper range.
For speech-focused or bandwidth-conscious encoding: a low-pass around 12–16 kHz can reduce high-frequency content while retaining substantially more naturalness than telephone-style speech. For a deliberately compact, voice-centred sound, an upper limit around 7–8 kHz is comparable to the bandwidth used by wideband speech systems, but it will sound less open than a normal podcast master.
For noisy archival narration or storytelling: try a high-pass around 80–125 Hz and, only if high-frequency hiss is objectionable, gradually lower the low-pass while previewing the result. Avoid using a narrow speech band unless intelligibility and file efficiency matter more than natural timbre.
These are starting points rather than fixed rules. Filter cutoffs are transition points, not brick walls, and the best values depend on the voice, microphone, background noise, output bitrate and intended sound. For music or mixed music-and-speech material, LAME's automatic filtering is normally the safer default.