Can your design [ Av = 2,000x, DC--20KHz +-0.1dB, SNR = 120dB (20 bit floor) ] be performed with one OpAmp? Is this for 20 or 24 bit audio?
edit[can that one opamp drive the ADC input charge demands of the ADC sample-hold circuit, yet settle very quickly?]
simulate this circuit – Schematic created using CircuitLab
First, what Rnoise is needed? With 120dB noise below 4 milliVolts RMS, you need 4 nanoVolts TOTAL INTEGRATED INPUT-REFERRED NOISE. That is, the noise must be 10^-6 smaller than the minimum input level; 4mV * 1e-6 = 4 nanoVolts RMS. In 20KHz bandwidth. To compute the Rnoise (sum of all random noise contributors in that first stage), divide the total integrated input-referred noise by squareroot of the bandwidth thus: 4nV/sqrt(20,000) = 4nV/141 = 30 picoVolt noise density per root hertz. With 66 Ohms Rnoise producing noise density of 1nanovolt/rtHz, and 66 milliOhms Rnoise (yes, << one ohm) producing 1nV/sqrt(1,000) = 33 picoVolts, you cannot achieve 120dB SNR with only 4 milliVolts RMS input signal. Why? the lowest OpAmp Rnoise is about 10 ohms, and more usually 50 ohms; the external gain-set resistors must be rather large [>>> 66 milliOhms, to avoid thermal distortion; even so, you'll need to include output buffers after the opamp, to avoid thermal distortion].
Now about the UnityGainBandWidth of the OpAmp: you'll need F3dB of approximately 200KHz to have 20KHz +-0.1dB. And you want a precision gain of 2,000X. The UGBW is F3dB * Av = 200,000Hz * 2,000 = 400,000,000. Using an opamp with that high a UGBW is a huge challenge.
If you want stereo imaging, you'll need matched left-right channel gain/phase, thus your opamps need enough excess gain to precisely control the gain/phase up to 20,000Hz. Precise? 0.1dB? that places the F3dB at 200,000Hz. A 10MHz UGBW opamp allows gain of 10,000,000 / 200,000 = 50X.
What is reasonable to attempt? Multi-stage opamp signal chain; first opamp with Rnoise of 50 or 60 ohms and UGBW of 10MHz; you'll need 50m * 50X = 2.5 volts RMS output at 20KHz. SlewRate is 2.5*1.414 *20,000 * 6.28 = 500,000 volts/second. From that first opamp.
In between first and second opamp, you'll need some type of variable attenuator, aka volume-control.
The 2nd opamp may be same as the first, with 15 volts/uS slewrate minimum.
Walt Jung has advice on picking opamps to provide low distortion at high slewrates.
Result? precision gain of 2,500x or 2,000x; SNR of 4mV/(1nV * sqrt(20,000) or 4mVolt/141nanoVolt or 28,000 (89dB SNR). SlewRate distortion is up to you.
If this 8 volts RMS output needs to drive an ADC, that ADC will demand sampling charge surges, and the OpAmp will need to SETTLE back to baseline voltage on the order of 0.1uSecond. The sampling charge surges will glitch the VDD filters, and cause ringing. You won't want to amplify that ringing, thus THREE opamps seem appropriate.