I'm unclear about what you mean exactly. By 96Khz, do you mean 96Khz sampling rate? Or do you mean a bitrate, implying some compression. You can use sox to do sample conversion using any of three different techniques. Your given "rate" figures look a bit strange... 5000.66Khz?! Please be explicitly clear about what you're talking about. The only commonly used sampling rates I've heard of are (all in KHz): 8, 11.025, 12, 16, 22.05, 24, 32, 44.1, 48, 96. I don't know what 89, 43.2, 92, or 5,000.66KHz would be useful for. As for bitrate selections, you have more choices with "MPEG 2" Official "ISO-approved" bitrates are as follows: For MPEG1 (sampling frequencies of 32, 44.1 and 48 kHz) n = 32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 320 For MPEG2 (sampling frequencies of 16, 22.05 and 24 kHz) n = 8, 16, 24, 32, 40, 48, 56, 64, 80, 96, 112, 128, 144, 160 For what it's worth, I've found the following LAME encoder settings to produce optimally sized and perfect-quality speaker recognition compressions of phone recordings: #!/bin/bash declare -i newscaleint newscale=`sox "$2.wav" -e stat -v 2>&1 | cut -b1-4` newscaleint=`echo ${newscale} | cut -f1 -d\.` if [ ${newscaleint} -gt 20 ] ; then newscale="20.0"; fi if [ "${newscale}" == "1.00" ] ; then { lame -q 2 -h --vbr-new -b 8 -B 32 "$2.wav" "$2.mp3" } else { lame -q 2 -h --vbr-new --scale $newscale -b 8 -B 32 "$2.wav" "$2.mp3" } fi You can skip the sox pass if you don't care about normalising the samples. =MB= -- A focus on Quality.