The accuracy of Frequency Analysis amplitude

Are the amplitude showed when we analyze an audio clip with Audacity Frequency Analysis accurate? How do we find the accurate amplitude when we’re using the Frequency Analysis?

Are you referring to “Plot Spectrum”?

If so, then a continuous sine tone with an amplitude the full height of a track will display as (about) 0 dB.
The display graph is an average for the selected audio, so the amplitudes at each frequency shown on the graph relate to the average level of that frequency within the selection. For example, if the selection comprises of a full-scale sine wave for half the duration, and silence for the other half, the spectrum will show the level as around -3 dB as that is the average level.

Yes, the Plot Spectrum.

I opened an audio clip which demonstrating a distortion pedal. The first thing I always do everytime I run clips with Audacity, I will select the entire audio and then analyze it with the Plot Spectrum (If the duration is beyond 4 minutes or so the Plot Spectrum will only analyze half of the entire audio). The amplitude keep changing everytime I change the size and and the axis so it is very confusing, for example the first time I checked the selected audio with the Plot Spectrum, the decibel level of the audio is say -48 Hz but then when I maximize the Plot Spectrum the amplitude change. Then, I change the size and the amplitude is changing again. Is it possible to see the accurate amplitude of clips with the Plot Spectrum? Anyway, the duration of the clip is 3 minutes 5 seconds, so which size and axis should I choose to improve the accuracy?

What is the suitable size and axis for 44100 Hz sample rate? Beyond 16384?

Are you referring to the “Size” parameter, or the physical dimensions of the effect window?

Assuming that you mean the “Size” parameter, it is normal and correct that most of the dB values will be lower with higher “Size” settings.
How it works:
Plot Spectrum performs “FFT analysis”, which in effects splits the audio into many “frequency bins”. For technical reasons (because it’s how FFT works), the number of bins is half of the “Size” setting. For example, with a Size of 1024, there will be 512 frequency bins, whereas with a Size of 2048 there will be 1024 frequency bin.
Now imagine that we are analyzing white noise, where there is approximately the same amount of sound across all audio frequencies. If there are 512 bins, then each “bin” will capture 1/512 of the sound. If on the other hand there are 1024 “bins”, the same white noise would be shared out between each bin and only half as much (1/1024) in each bin. The spectrum graph indicates how much is in each bin.

Yeah, I’m talking about the Size where you can choose from 128 to 65536.

In order to find the most accurate amplitude, which Size should I choose? The audio clip is around 4 minutes in length and the sample rate is 44.1khz. Also, which axis is more accurate when looking at amplitude? Is it Log or Linear?

The important thing is that when comparing audio, you use the same “Size” parameter.

I assumed you are talking about comparing two or more different audio clips by setting them with the same size parameter. Yes, I’m actually trying to compare the amplitude of an audio to the amplitude of different audios. Thanks.