Skip to the content.

This is the accompanying page for the article Dequantized signal from two parallel quantized observations authored by Vojtěch Kovanda, submitted to the student conference EEICT 2025.

We propose a technique for signal acquisition that uses a combination of two devices with different sampling rates and quantization accuracies. Subsequent processing involving sparsity-based regularization enables us to reconstruct the input signal at such a sampling frequency and with such a bit depth that were not possible using the two devices independently. Objective tests show the evaluation of the proposed method in comparison with the alternatives.

Audio examples

You can listen to selected audio excerpts. Other examples can be generated by running the available Matlab code.

Tables below show the effective number of bits per sample, allowing fair comparison between methods. For example, the case y1 and y2 quantized to 12 and 4 bps, respectively, corresponds to 7 effective bits per sample. For 8 bps, we provide examples of quantized audio using iZotope's MBIT+ dithering algorithm as another method for removing quantization noise.

Clarinet, 7 bps

Original audio (x) 24 bps
Quantized audio (xQ) 7 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 7 bps
Our reconstruction from y1 and y2 7 bps

Violin, 7 bps

Original audio (x) 24 bps
Quantized audio (xQ) 7 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 7 bps
Our reconstruction from y1 and y2 7 bps

Flute, 7 bps

Original audio (x) 24 bps
Quantized audio (xQ) 7 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 7 bps
Our reconstruction from y1 and y2 7 bps

Comparison of 7 bps audio in terms of SDR and ODG

Image 1 Image 2

Clarinet, 8 bps

Original audio (x) 24 bps
Quantized audio (xQ) 8 bps
Dithered audio (iZotope's MBIT+) 8 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 8 bps
Our reconstruction from y1 and y2 8 bps

Violin, 8 bps

Original audio (x) 24 bps
Quantized audio (xQ) 8 bps
Dithered audio (iZotope's MBIT+) 8 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 8 bps
Our reconstruction from y1 and y2 8 bps

Flute, 8 bps

Original audio (x) 24 bps
Quantized audio (xQ) 8 bps
Dithered audio (iZotope's MBIT+) 8 bps
Reconst. via the CP algorithm, i.e. CP(xQ) 8 bps
Our reconstruction from y1 and y2 8 bps

Comparison of 8 bps audio in terms of SDR and ODG

Image 1 Image 2