>>19401101(in the past, for really broken stuff, i would sometimes use the karaoke track to isolate her vocals and then reassemble a new version of it using the karaoke track instead of her mix, but that didn't work starting from october and i haven't yet got to the streams after she ditched the global EQ).
8. try to get out of the end of the track quickly and smoothly. if her reverb is on, look for a reverb tail that she lets run out almost to silence and trim it and use that. if there isn't one, grab the last bit of talking you want, feed it into a new 100% wet reverb, cut the talking, crossfade the reverb tails, and adjust the level to match.
9. leave 2 seconds of silence at the end of the track. export to 32-bit WAV at whatever sample rate (i.e. 48 kHz for archived opus, 44.1 kHz for unarchived AAC).
after you have all the WAVs:
10. normalize the whole collection of WAVs. after cutting out all the chatting sections from the raw stream audio, and fixing gain changes and compressor attack spikes, you'll always end up peaking lower than 0 dBFS. the standard advice here is to normalize to -1 dBFS, but i usually aim for -0.25 dB. my goal is to be at or above the replaygain loudness target.
11. export the normalized WAVs in 16 bit format. encode to FLAC and throw away the 16-bit WAVs.
12. tag the FLACs. resize the stream thumbnail to 600x338 and save it as a baseline (not progressive) JPEG and tag the files with it as the front cover. save the original thumbnail as cover.jpeg.
13. encode to AAC (qaac --tvbr 109 -q 2 -r keep) and MP3 (lame -q0 -V1.25).
14. generate replaygain tags for all the files. if you are under the target (i.e. album gain is positive) consider more aggressive limited on problem files, re-export and return to step 10.
15. fill out documentation and upload.
