@ RlyDontKnow:
Therefore define "a lot"...
"Audibly transparent quality" means that a majority of listeners could not tell apart the original from the source in an "ABX listening test".
The transparency threshold was tested by members of the HydrogenAudio forum (namely Roberto Amorim), where technically educated people meet to optimize codecs. They know audio compression algorithms in detail.
Of course, all psychoacoustic algorithms decrease objective quality. There is only a decision how much decrease is noticable or acceptable. For LAME, the averagely noticable threshold is near the result of the VBR preset "standard", so the preset "extreme" has a lot of headroom even for audiophile listeners. For Ogg Vorbis, the averagely noticable threshold may be around VBR quality level 4..6 (to achieve comparable results to LAME MP3 presets standard..extreme). If you feel a bit paranoid, you can as well ask for Ogg Vorbis quality level 10, which is objectively (measurably) even better than the MP3 algorithm is even able to store at constant 320 kbps (which would be a complete waste on human speech anyway). And if you demand lossless quality, then allow at least lossless compressors like WavPack, Monkey's Audio, FLAC etc.
But 96 kHz? Did you ever take a look at the frequency response curves of average consumer microphones sold in music stores, or even worse, built-in in headsets? I don't know if there are consumer microphones with a guaranteed frequency range above 22 kHz. Furthermore: The whole-spectrum frequency loss during the recording of your speech is by magnitudes worse than the frequency loss of tuned psycho-acoustic algorithms in quality levels mentioned above. Don't be afraid of multiplying loss ... I would agree for CBR 128 kbps MP3 created by any average encoder, but not for LAME VBR preset extreme or any encoder better than MP3 with at least "HydrogenAudio certified" transparent quality level.
But 32 bit? Do you know any PC soundcard which is technically, physically able to sample in a finer resolution than 16 bit from the microphone input? You won't find those on-board. You won't find those among "gamer" soundcards. I am not even certain if any musician sampling soundcard would provide 24-bit sampling via microphone input - if at all, then via line-in, so you would need a mixer in front of it.
And then we shall upload uncompressed samples ... even 7-zip compressed (which is highly inefficient anyway due to half of the sample resolution being rather random junk than exact information) it is a waste of bandwidth. As uncompressed samples with 32 bit and 96 kHz (which is interpolated in the majority, not recorded) there are 5.76 Megabytes per recorded minute, multiplied by up to 5 takes per sentence so you can later chose the best take), multiplied by the number of sencentes. You will easily collect hundreds of Megabytes. And uploading those may take several hours, depending on your upload bandwidth (which is a tenth your download bandwidth for most ADSL connections).
You want more than 400% of the quality your hardware is even technically able to record. And people wouldn't notice any loss if I sent you just 40% of that level.
My file is bigger. Fapp. Who is Claude Shannon? What is Entropy in a context of Information? Never heard of.
(No offense. Just sense.)
__
@ Talad:
... So I may need a new sound card to record more. (Leaving the cable in the plug for months must have destroyed my current one, possibly electrostatically.)
