Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec (Read 46715 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #100 –
Quote from: genuine on
Nowadays, there is no place for low and very low quality audios. Instead, it would be more logical to work for lossy but higher quality (320 kbps+) results.
(Quality is not measured in bits/s)
but I agree with the point that perhaps the most interesting goal is to achieve transparency at less bits/s compared to other codecs which can do that, and also
* with less possible "killer samples" ("non-musical" and "synthetic" or not), ideally none
* covering the abilities of "golden ears" as well (not just "average" listener), or at least easily tunable to do so
* bonus point, somehow provable that this would always apply for a given human hearing model, not just relying on a finite number of test cases × a finite number of listening test participants
a fan of AutoEq + Meier Crossfeed

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #101 –
Hello everyone. I hope soon to publish the latest version of ADC which, as mentioned a few days ago, is still time-dominated. With the next version I will introduce fast and slow compression. Currently the minimum compression value is around 256 kbps and with a quality (measured with Peaq parameters) comparable and often superior to Wavpack option "-b2.6 -hh". Of course, your listening is obviously missing.  It obviously cannot compete in terms of overall quality with frequency domain codecs.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #102 –
Can't wait!
 

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #103 –
Title: Update on ADC - Delay in Releasing the Next Version

Hello everyone,

I wanted to provide an update on ADC, as I know many have been waiting for a new version for the past few months. First of all, I appreciate your patience and interest in the project.

Over the last 4-5 months, I have conducted around 100 experiments to further refine ADC, and the results have exceeded my expectations. The improvements in distortion reduction, noise shaping, and overall transparency at lower bitrates have reached a point where ADC is performing significantly better than expected—often surpassing well-established codecs in direct comparisons.

Given these results, I need to reconsider my approach to distribution. Initially, I had planned to release a free, unrestricted version, but given the current performance of ADC, I have to reevaluate the best path forward before making it publicly available.

This does not mean I am abandoning the idea of a public release, but rather that I need more time to determine the right balance between sharing my work and ensuring that the project continues to evolve in the best way possible.

I hope this clarifies the delay, and I appreciate your understanding. Feel free to ask any questions!

Best, Nania Francesco

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #104 –
Quote from: Nania Francesco on
My system doesn't use mdct- fft - wavelet transformations (like most lossy codecs) .
Nania; what kind of approach do you use roughly in ADC? I mean, do you work in time domain or frequency domain? Or hybrid?

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #105 –
I've made significant improvements to the latest version of the ADC codec. Compared to version 0.70, the enhancements are around 80% in terms of efficiency: I'm now able to achieve the same PEAQ audio quality with better technical parameters.

For example, I was able to reduce the bitrate from 306 kbps to 256 kbps while still obtaining slightly better results than the previous version. This shows that the codec is now more efficient without compromising quality.

I'll be sharing some short comparison demos as soon as possible.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #106 –
Quote from: genuine on
Quote from: Nania Francesco on
My system doesn't use mdct- fft - wavelet transformations (like most lossy codecs) .
Nania; what kind of approach do you use roughly in ADC? I mean, do you work in time domain or frequency domain? Or hybrid?
One of the key innovations of my codec lies in how it leverages the time domain—in a way that, to my knowledge, no other codec has fully explored before. Instead of relying on traditional transformations into the frequency domain (like FFT or MDCT), this approach works directly on the waveform, preserving its structure and dynamics with minimal complexity.

At this stage, the codec does not apply any frequency-based filtering, psychoacoustic masking, or band cutting techniques. Despite this, it achieves impressive results purely through efficient time-domain modeling and signal adaptation.

This makes the codec both lightweight and highly transparent, especially at medium-to-low bitrates.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #107 –
I' ve prepared three audio compression demos to showcase the progress of my codec:

ADC 0.70 – the previous stable version

Current ADC (work-in-progress) – an improved but still non-final build

WavPack – used here as a reference at a similar bitrate

The goal is to provide a fair comparison at equivalent bitrates, highlighting the evolution of quality and the efficiency of the new approach.

My codec is designed with a unique time-domain-centric model that does not rely on frequency filtering or masking, yet manages to preserve clarity and dynamics. The latest version brings around 80-90% improvement in efficiency (
understood as a reduction of the bitrate with the same efficiency !) : for example, it achieves better PEAQ scores than version 0.70, at a lower bitrate (e.g., from 306 kbps down to 256 kbps), showing clear gains in both transparency and compression.
The 29-second demo is available free online.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #108 –
New updates

Introducing 24-bit Support for ADC – High-Resolution Audio, Even at Low Bitrates! 🎧

I'm excited to share that my custom audio codec ADC now supports 24-bit WAV input, bringing high-resolution audio capabilities to the next level. 💡

In recent tests, ADC outperforms WavPack at low bitrate settings (e.g., ~220-256 kbps), achieving:

Higher PEAQ quality scores

Lower distortion levels

Better noise performance

More efficient time-domain handling

⚙️ Unlike traditional codecs, ADC leverages the time domain in a unique way—no frequency cutting, no pre-filters. Just clean, dynamic compression tailored for precision.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #109 –
Cool, but still no binaries.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #110 –
🎧 New ADC Codec Demo Available – Up to 31 Seconds of Audio Compression 🚀

 I’m releasing a demo version of my ADC audio codec, allowing maximum compression tests on audio clips up to 31 seconds.

This is a chance to experience the codec's behavior and quality at different bitrates — especially in low-bitrate scenarios, where ADC shows significant advantages compared to traditional solutions like WavPack.

🧪 Features:

Support for 16, 24 WAV input

Unique time-domain-based compression (no frequency cutting, no filters)

Designed to preserve clarity and dynamics even under heavy compression

Feel free to test and share your feedback — more improvements and tools are on the way!

🎧 8 Months of Work – A Pure Time-Domain Audio Codec

After more than 8 months of continuous development, I feel confident saying that the results of my custom audio codec are genuine and promising.

It’s true — frequency-domain and hybrid codecs have achieved impressive results by masking certain artifacts and perceptual limitations. Techniques like psychoacoustic modeling, joint stereo, pre-echo smoothing, and frequency band cutting have become standard practices in audio compression.

But my goal has always been different.

🔍 I believe that when we compress a sound, we should aim to preserve its original characteristics as faithfully as possible — without relying on masking, tricks, or perceptual compromises. That means:

No joint stereo

No pre-echo compensation

No frequency trimming

No post-processing “patches”

Just pure time-domain compression, driven by intelligent data analysis and structural preservation.

I know this is a challenging path, but I also believe it opens doors to new kinds of fidelity — especially in low-bitrate scenarios where traditional methods start to fall apart.

Thank you to everyone who has supported and tested the codec so far. More demos and technical breakdowns are coming soon.

I fixed the settings that could cause system crashes. The -q option goes from 20 to 511 . If you enable -h it will slow down but improve the final compression with the improved peaq values.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #111 –
I've recently compared ADC with WavPack at 352.8 kHz sample rate.
While WavPack produces a clean result, it achieves it by discarding many high-frequency details — leading to a PSNR around 35.6 dB.
On the other hand, ADC preserves the original waveform more accurately, resulting in a PSNR of 42.8 dB, with more harmonic integrity at the cost of occasional high-frequency instability.

This proves how ADC handles time-domain fidelity exceptionally well — especially for high-resolution audio. Future versions will address the fine transients that appear at extreme sample rates without compromising precision.
ADC vs WavPack at 88.2 kHz – A Closer Look

After extensive testing, ADC achieves superior perceptual quality compared to WavPack at similar bitrates (382 kbps vs 400 kbps). While WavPack slightly extends bandwidth, ADC delivers.

This test confirms that ADC’s time-domain approach preserves audio integrity more naturally, especially at high sample rates, without relying on psychoacoustic masking or aggressive spectral cuts. Naturally at the moment the most used frequency rarely exceeds 48khz

Next step: optimizing transient representation and adaptive shaping!

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #112 –
Hi everyone,

About a week ago, I posted a demo of my custom audio codec (max duration 31 seconds), and unlike in the past, I haven’t received much feedback this time around. No critiques, no bugs reported — which could be a good sign, or maybe it just flew under the radar!

Either way, I wanted to bring it back to your attention, as I’ve made a few minor improvements since the last version — mostly around quality consistency and a bit of refinement in the decoder logic.

If you haven’t tried it yet, I’d really appreciate it if you could give it a listen and let me know what you think — whether you like it, dislike it, or think it could be used in a specific context. I’m still exploring where this codec fits best (e.g., games, embedded systems, retro-style audio, etc.).

Feel free to test it with any short sample (up to 31 seconds), and let me know how it sounds on your setup!

Thanks for checking it out 🙏

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #113 –
Just tested, with -t320 and with random audio FLAC sample down-mixed to mono, trimmed from start to 30 sec duration.
There is strange noisy sound at start of audio that is not present in source.
The high-end spectrum content is too much distorted for -t320 IMHO.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #114 –
Quote from: mycroft on
Just tested, with -t320 and with random audio FLAC sample down-mixed to mono, trimmed from start to 30 sec duration.
There is strange noisy sound at start of audio that is not present in source.
The high-end spectrum content is too much distorted for -t320 IMHO.
the -t320 option is not to be used because the -t options range from -t0 , -t1 .... -tx at most and have a function only with -h active to apply a sort of filter with t0 on a large range, with tx on a small data range. If you put the test sound online I can verify.  The factor that determines the quality is -q h and it ranges from -q20 to -q511 but these are quantization values ​​not kbps, in general your version with -q60 should produce around 300-320 kbps stereo and 160 kbps mono . The factor that determines the quality is -c h and it ranges from -q20 to -q511 but these are quantization values ​​not kbps, in general your version with -q60 should produce around 300-320 kbps stereo and 160 kbps mono   Another important thing. The audio to test must not contain images or other extra data because at the moment I have not implemented the recognition of extra data of the wav header. If it reads here data for him it is music and causes distortion. Thanks for the test

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #115 –
Even with -q240 the noisy artifact is still there, i made sure that input wav have 0 metadata stored.

Here is that input file:

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #116 –
I did some listening tests and at least I don't perceive any difference with the original in the initial part. However I did a more in-depth analysis with a program and something is noticeable at the delta level, but it is found in all time domain codecs, some more, some less.  Like all time domain codecs, the more you advance the compression, the fewer imperfections there are.Anyway, thanks for the detail.

Re: ADC (Adaptive Differential Coding) My Experimental Lossy Audio Codec

Reply #117 –
You need to pause after start of playing and make sure that DSP is not triggering fade in/out.

Its very obvious to spot when listening.