Isn't the lossless compression method more noise-resistant?

Hi everyone

I do an audio compression with bit per sample variations

The original file is .wav (2 channels) and compressed using the lossless .flac method with variations in bits per sample (24,16, and 8)

When measuring objective difference grade (ODG) on .flac audio with 16 and 24-bit samples compared to the original file there is no problem (ODG=0)

However, when checking the audio .flac bitpersample 8 compared to the original file the result becomes -3 (Annoying)

Why can this happen?

Even though I used a lossless method which should be more resistant to noise

Isn't the lossless compression method more noise-resistant?

I'm sorry, i cant speak english very well

Thank you very much

Answer 1

[Image Analyst]

If your original (best) signal has 24 bits per sample then quantizing it to 8 or 16 bits per sample will degrade the waveform. The degraded waveform will have some difference from the original. I don't know that I'd call it "noise" exactly. Rather than an alteration of the signal due to a random process, it's more of a quantization error due to a systematic process. Even though the signal is written and recalled with a noiseless compression, there will be a difference between the quantized signal and the original due to quantization.

In general lossless compression is completely "resistant" to noise in that if you save something with lossless compression and then read it back in, it will be the same as what you originally saved.

If you have lossy compression, the recalled signal will not be the same as the original. We usually call the differences "compression artifacts" rather than "noise."

[Sisi Misi]

is there a way to preserve the compressed form of the signal?

To get the ODG close to 0

especially when compressing .wav audio to .flac with 8 bitpersample variation

[Image Analyst]

I'm not sure what you mean. Lossless compression will conserve the signal in that it will be the same after a round trip compress, save, recall, uncompress. If you reduce the number of bits per sample, or if you use lossy compression, obviously you are throwing away information and there is not way to get it back 100% afterwards. You could try some interpolation but it won't be exact for every sample point.

[Sisi Misi]

Ok Mr, thank you

But, based on your answer

The audio compression process with variations in bits per sample will degrade the waveform or a quantization error due to a systematic process.

And I have quastion about it

My first question

How can we reduce the quantization error in the lossless compression process with this bit-persample variation method?

Second question, is there any other way to do lossless audio compression in matlab with no errors or according to lossless compression theory?

[Image Analyst]

I'm not an expert in different ways (formats) to compress audio with loss to reduce the loss. I don't even know what formats are lossless or lossy so you know more than me already. Sorry - my expertise is in image processing not audio signal processing.

[Sisi Misi]

alright Mr

Thank you so much