After a lot of hard work I managed to get this version 0.70 which I hope you like more than the previous ones. I managed to insert a CBR/ABR style bitrate control which certainly needs improvement but which does its job well.

I hope you can advise me. Greetings!

only from:
http://heartofcomp.altervista.org/ADCodec.htm

After a lot of hard work I managed to get this version 0.70 which I hope you like more than the previous ones. I managed to insert a CBR/ABR style bitrate control which certainly needs improvement but which does its job well.

I hope you can advise me. Greetings!

only from:
http://heartofcomp.altervista.org/ADCodec.htm

All those measurements are irrelevant with perceptual lossy codecs. Only thing that matters is how they sound.

Edit: clarification: All those measurements are *wrong* for perceptual lossy codecs. They don't measure audio the way ear hears things.

I thank everyone for the tests (except for someone who doesn't deserve respect for the ridiculous comments he makes). For the -b26 option it makes no sense as at minimum it is set to 128 kbps but being an 8 khz sound which is already terrible in quality. I respect the choice of tests which is considerably reductive. In my tests on "Sopranino Recorder Concerto, RV 443_ Allegro" by Vivaldi test file for example with EAQUAL - Evaluation of Audio Quality ( https://github.com/spxnn/eaqual - https://www.rarewares.org/files/others/eaqual.zip ),

mp3 192 kbps  results:
Resulting ODG:  -0.16
Resulting DIX:   2.32
BandwidthRef    17511.4298
BandwidthTest   15909.6910
NMR                 -15.6350
WinModDiff1     5.0334
ADB                   -0.8332
EHS                   0.2262
AvgModDiff1     5.1856
AvgModDiff2     9.0777
NoiseLoud         0.0853
MFPD                  1.0000
RDF             0.0041

while with ADC always at 192 kbps I get this:
Resulting ODG:  -3.05
Resulting DIX:  -1.26
BandwidthRef    11930.2636
BandwidthTest   11913.7601
NMR                  -2.9485
WinModDiff1     12.0720
ADB                  1.8283
EHS                    0.2967
AvgModDiff1     10.4023
AvgModDiff2     26.8317
NoiseLoud        0.3622
MFPD               1.0000
RDF                  0.7832

There is certainly a gap with the management of CBR/ABR. But ADC with the options -tx -q12  I get for example:
Resulting ODG:  -0.88
Resulting DIX:   1.03
BandwidthRef    14253.3514
BandwidthTest   14252.6861
NMR             -13.1225
WinModDiff1     4.6411
ADB             1.3101
EHS             0.2593
AvgModDiff1     3.6089
AvgModDiff2     8.0902
NoiseLoud       0.1116
MFPD            1.0000
RDF             0.0512


ODG (Objective Difference Grade): Measures perceived fidelity, where scores range from 0 (perfect fidelity) to lower negative values indicating increased perceptible quality degradation. A higher (closer to zero) ODG indicates better quality.

DIX (Distortion Index): Represents the total distortion level within the processed audio, with lower values indicating better quality. It's a metric that combines several distortion measures to offer an overall indication of the codec's accuracy.

BandwidthRef and BandwidthTest:

BandwidthRef is the reference (original) signal's bandwidth in Hz, indicating the highest frequency component captured in the original audio.
BandwidthTest is the bandwidth after compression and decompression, showing the highest frequency maintained post-processing. Lower values suggest possible loss of high-frequency information.
NMR (Noise-to-Mask Ratio): Reflects the audibility of noise introduced by compression. A more negative NMR value indicates that the added noise is less perceptible because it’s effectively masked by the audio signal, leading to a cleaner, more faithful reproduction.

WinModDiff1: Represents the windowed difference in signal modulation (amplitude/phase distortion) over short windows, where lower values indicate a more faithful signal with minimal short-term distortion.

ADB (Average Distortion per Band): Measures the average distortion within distinct frequency bands. Lower ADB indicates more accurate reproduction across the audio spectrum.

EHS (Error Harmonic Structure): This parameter measures the harmonic structure of the error or distortion introduced, with lower values suggesting that the distortion is less likely to be perceived as unnatural by listeners.

AvgModDiff1 and AvgModDiff2:

AvgModDiff1 is the average modulation difference over short time scales, indicating how closely the processed signal matches the original in terms of amplitude and phase.
AvgModDiff2 is the long-term version of AvgModDiff1, measuring modulation similarity over longer time windows.
NoiseLoud (Noise Loudness): Measures the perceived loudness of noise introduced by the codec. Lower values indicate that the noise is less noticeable to the human ear, leading to a cleaner sound.

MFPD (Mean Frequency Perceptual Deviation): Measures the perceptual accuracy of high frequencies. A perfect score of 1.0 indicates no deviation from the reference signal in high-frequency perception.

RDF (Relative Delay Factor): Indicates any phase or timing discrepancies between the reference and test signals. Lower values suggest minimal delay, which helps maintain a more natural and cohesive audio reproduction.
I don't give up on what you say. I'm not looking for innovation but a different way to compress audio.

Quote from: Case on 2024-11-03 18:32:25

All those measurements are irrelevant with perceptual lossy codecs. Only thing that matters is how they sound.

Edit: clarification: All those measurements are *wrong* for perceptual lossy codecs. They don't measure audio the way ear hears things.

Yeah. With hearing I found serious problems with lower (and default q4) presets. Only higher presets can give decent quality but compression is quite poor. No idea if it's okay compared to ADPCM, but at least seems similar (even worse?) than QOA.
This is valid only for version 0.60 since I didn't test 0.70, but I don't expect great improvements at this point.

Quote from: Nania Francesco on 2024-11-03 18:13:39

I thank everyone for the tests (except for someone who doesn't deserve respect for the ridiculous comments he makes). For the -b26 option it makes no sense as at minimum it is set to 128 kbps but being an 8 khz sound which is already terrible in quality. I respect the choice of tests which is considerably reductive. In my tests on "Sopranino Recorder Concerto, RV 443_ Allegro" by Vivaldi test file for example with EAQUAL - Evaluation of Audio Quality ( https://github.com/spxnn/eaqual - https://www.rarewares.org/files/others/eaqual.zip ),

mp3 192 kbps  results:
Resulting ODG:  -0.16
Resulting DIX:   2.32
BandwidthRef    17511.4298
BandwidthTest   15909.6910
NMR                 -15.6350
WinModDiff1     5.0334
ADB                   -0.8332
EHS                   0.2262
AvgModDiff1     5.1856
AvgModDiff2     9.0777
NoiseLoud         0.0853
MFPD                  1.0000
RDF             0.0041

while with ADC always at 192 kbps I get this:
Resulting ODG:  -3.05
Resulting DIX:  -1.26
BandwidthRef    11930.2636
BandwidthTest   11913.7601
NMR                  -2.9485
WinModDiff1     12.0720
ADB                  1.8283
EHS                    0.2967
AvgModDiff1     10.4023
AvgModDiff2     26.8317
NoiseLoud        0.3622
MFPD               1.0000
RDF                  0.7832

There is certainly a gap with the management of CBR/ABR. But ADC with the options -tx -q12  I get for example:
Resulting ODG:  -0.88
Resulting DIX:   1.03
BandwidthRef    14253.3514
BandwidthTest   14252.6861
NMR             -13.1225
WinModDiff1     4.6411
ADB             1.3101
EHS             0.2593
AvgModDiff1     3.6089
AvgModDiff2     8.0902
NoiseLoud       0.1116
MFPD            1.0000
RDF             0.0512


ODG (Objective Difference Grade): Measures perceived fidelity, where scores range from 0 (perfect fidelity) to lower negative values indicating increased perceptible quality degradation. A higher (closer to zero) ODG indicates better quality.

DIX (Distortion Index): Represents the total distortion level within the processed audio, with lower values indicating better quality. It's a metric that combines several distortion measures to offer an overall indication of the codec's accuracy.

BandwidthRef and BandwidthTest:

BandwidthRef is the reference (original) signal's bandwidth in Hz, indicating the highest frequency component captured in the original audio.
BandwidthTest is the bandwidth after compression and decompression, showing the highest frequency maintained post-processing. Lower values suggest possible loss of high-frequency information.
NMR (Noise-to-Mask Ratio): Reflects the audibility of noise introduced by compression. A more negative NMR value indicates that the added noise is less perceptible because it’s effectively masked by the audio signal, leading to a cleaner, more faithful reproduction.

WinModDiff1: Represents the windowed difference in signal modulation (amplitude/phase distortion) over short windows, where lower values indicate a more faithful signal with minimal short-term distortion.

ADB (Average Distortion per Band): Measures the average distortion within distinct frequency bands. Lower ADB indicates more accurate reproduction across the audio spectrum.

EHS (Error Harmonic Structure): This parameter measures the harmonic structure of the error or distortion introduced, with lower values suggesting that the distortion is less likely to be perceived as unnatural by listeners.

AvgModDiff1 and AvgModDiff2:

AvgModDiff1 is the average modulation difference over short time scales, indicating how closely the processed signal matches the original in terms of amplitude and phase.
AvgModDiff2 is the long-term version of AvgModDiff1, measuring modulation similarity over longer time windows.
NoiseLoud (Noise Loudness): Measures the perceived loudness of noise introduced by the codec. Lower values indicate that the noise is less noticeable to the human ear, leading to a cleaner sound.

MFPD (Mean Frequency Perceptual Deviation): Measures the perceptual accuracy of high frequencies. A perfect score of 1.0 indicates no deviation from the reference signal in high-frequency perception.

RDF (Relative Delay Factor): Indicates any phase or timing discrepancies between the reference and test signals. Lower values suggest minimal delay, which helps maintain a more natural and cohesive audio reproduction.
I don't give up on what you say. I'm not looking for innovation but a different way to compress audio.

thanks for providing link to EAQUAL will closely inspect it. (The project seems last active 7 years ago)

Also you could try other audio evaluation scores like PSNR SDR SISDR MAE NRMSE MDA all available in librempeg.

I managed to find an exe version of PQevalAudio v3r0. I retested the wav file I mentioned at 48 khz and obtained these results which confirm what you said, I don't know if in part.

If you are a C++ programmer, download mp3 lame and see how many dozens of filters there are to extrapolate and cut. purify and clean up the original sound which will obviously be artificial and artificial. A summary. Needless to say, most codecs produce artificial, high-quality clipped sound. Simple.