Ok. Let's start from the very beginning.
First, noise floor in terms of power/sqrt(hz) is what you need to know. You have to calculate that from the 16 bit floor. Now you know energy per sqrt(hz). Easiest to assume +-1 is maximum level when you do this, and keep units arbitrary and relative to +-1.
Then you need to calculate the energy (sum across frequency in Hz in energy) in an ERB. For simplicity, use 50Hz bandwidths at low frequencies and 1/4 Octave above the point where that's bigger than 50Hz. Now you know the ENERGY in that band, relative to the initial signal max of +-1.
Now you know the level IN THAT BAND. Since the noise is spread out, it will always be smaller than the total noise level.
THAT is the energy in a given ERB. That is the energy that reads on detectability of the noise by the ear, in that ERB.
Now, a sine wave will disappear, monophonic, at -5.5dB below that. At frequencies below 500Hz, a stereophonic signal with coherent sine wave, and incoherent noise in the two channels will get you down to almost -30dB below that level. This ability to detect the binary masking level difference will start to disappear above 500Hz, and be mostly gone by 2kHz. HOWEVER if your noise has a palpable envelope, i.e. it's not more or less constant level, then above 2kHz you can get unmasking from the noise envelope when listening binaurally, again maxing out either at absolute threshold (i.e. zero loudness level) or at about -30 dB relative to the noise. It is, however, rare for an input signal and noise generator to get there above 2kHz. You almost have to contrive it.
ALMOST.
Please refer to my talk on how the ear works to make sense of this. It's a 2 hour talk to technical people who are paying rapt attention.
First, noise floor in terms of power/sqrt(hz) is what you need to know. You have to calculate that from the 16 bit floor. Now you know energy per sqrt(hz). Easiest to assume +-1 is maximum level when you do this, and keep units arbitrary and relative to +-1.
Then you need to calculate the energy (sum across frequency in Hz in energy) in an ERB. For simplicity, use 50Hz bandwidths at low frequencies and 1/4 Octave above the point where that's bigger than 50Hz. Now you know the ENERGY in that band, relative to the initial signal max of +-1.
Now you know the level IN THAT BAND. Since the noise is spread out, it will always be smaller than the total noise level.
THAT is the energy in a given ERB. That is the energy that reads on detectability of the noise by the ear, in that ERB.
Now, a sine wave will disappear, monophonic, at -5.5dB below that. At frequencies below 500Hz, a stereophonic signal with coherent sine wave, and incoherent noise in the two channels will get you down to almost -30dB below that level. This ability to detect the binary masking level difference will start to disappear above 500Hz, and be mostly gone by 2kHz. HOWEVER if your noise has a palpable envelope, i.e. it's not more or less constant level, then above 2kHz you can get unmasking from the noise envelope when listening binaurally, again maxing out either at absolute threshold (i.e. zero loudness level) or at about -30 dB relative to the noise. It is, however, rare for an input signal and noise generator to get there above 2kHz. You almost have to contrive it.
ALMOST.
Please refer to my talk on how the ear works to make sense of this. It's a 2 hour talk to technical people who are paying rapt attention.
Last edited: