Measuring DC offset in an audio signal

Q 

I need to measure the DC offset of an audio signal at various points in a
circuit. The signal amplitude can vary between +-12V. Ideally, I would like to
be able to measure offsets with a precision of a few micro volts. I was
thinking of using an ADC like AD7794 to get the resolution and to to have a low
inherent offset, driven by an ADA4528-2 to get the correct amplitude at the
input of the ADC (between 0 and 5V, with no signal = 2.5V)
To get the best resolution using AD7794, I need to sample at about 4Hz. I
gather that I therefore need an antialiasing filter at around 2Hz or lower,
which would prove very difficult to implement. Needing the resistor values of
the filter to remain low ( because of the Johnson noise), I would need massive
capacitors, very impractical.
Do I really need to filter prior to the AD7794 (audio bandwidth is several
100kHz)? What would happen to the DC part of the signal if there was no filter?

Another method I can think of is to use an audio delta sigma ADC, measure the
whole audio bandwidth and filter in the digital domain (in a DSP), but these
audio ADC usually have poor DC offset characteristics, making it I believe
impossible to get an accurate DC offset measurement.

My last idea is to use an SAR ADC, but I am so far stuck with the resolution of
the SAR (16 bits max usually). With 5V Full scale, a 16 bit step is about 75uV
and I would ideally like a better accuracy. Would averaging in a DSP give me a
much better accuracy, I am not sure. Heavily filtering before the ADC and I am
stuck with the same problem as before.

Do you see other methods to accurately measure the DC offset of such an audio
signal?
Am I making a mistake in my 3 ideas above?

Please bear in mind that the ADC output will be connected to a 400 series Sharc
DSP, so there is some filtering / computing power available.

Your views are greatly appreciated.

 

A 

The attached Excel file shows the filter response of the AD7794 for an output
data rate of 4Hz. You should add the standard RC anti-aliasing filter that we
recommend which provides rejection at the modulator sampling frequency. This
anti-aliasing filter along with the AD7794's digital filter should reject the
audio tone and leave the DC signal. I do not think that a filter with a 3dB
point of a few Hz is needed. I have never tried this method before so I am not
guaranteeing it. However, in principle, it should work.