DSPi (Weeb Labs)
Open-source firmware that turns a Raspberry Pi Pico (RP2040) or Pico 2 (RP2350) into a USB sound
card with an onboard DSP engine. Covered by CNX Software (Jean-Luc Aufranc, 2026-10-07, tip from
Hedda); GitHub WeebLabs/DSPi, GPL-3.0, ~1.3k stars, 79 forks, 347 commits, 14 open issues. First
posted on the Audio Science Review forum in February 2026; that thread runs 121 pages. This is the
primary page for a project already named second-hand in audio-dsp-hardbag-guide, which quoted the
same channel/band/delay figures from a reseller's write-up without a project page to point to.
What it is
Both cores run overclocked to 307.2 MHz. Input is USB audio (16/24-bit, 44.1/48/96 kHz) or 24-bit S/PDIF, recognized as a standard audio device on macOS, Windows, Linux and iOS, with no driver install. Output is up to four stereo slots (S/PDIF or I2S) plus a mono PDM subwoofer output that skips a second DAC by using a 2nd-order delta-sigma modulator in firmware.
Processing, matching audio-dsp's five-function taxonomy: per-channel preamp; a matrix mixer with gain and phase invert at every crosspoint; up to 10 parametric EQ bands per channel (11 filter types); up to four crossover filters per output (Linkwitz-Riley, Butterworth or Bessel, up to 8th order); an upward-compressing leveller; ISO 226:2003 loudness compensation; BS2B headphone crossfeed; and per-output delay up to 85 ms (4,096 samples at 48 kHz). Control is through a GUI (DSPi Console) or a CLI (DSPi Terminal), both cross-platform; firmware updates over USB without pressing BOOTSEL.
RP2040 vs RP2350
| RP2040 | RP2350 | |
|---|---|---|
| Arithmetic | Q28 fixed-point, hand-written ARM assembly | single-precision float, hardware FPU |
| EQ bands (total) | 70 | 110 |
| Channels | 7 | 11 |
| Output slots | 2 stereo | 4 stereo |
| Mixer | 2×5 | 2×9 |
The 110-band, 11-channel, 85 ms figures already held in audio-dsp-hardbag-guide are the RP2350 ceiling, not a baseline. The RP2040 version is a smaller machine on fixed-point math.
Where it sits next to CamillaDSP
camilladsp is the spoke's other open DSP engine, and the two sit at opposite ends of the silicon this corpus tracks: CamillaDSP runs on a Raspberry Pi 5 (a Linux SBC, Rust engine, float throughout) as a RASPIAUDIO appliance; DSPi runs on a $4–5 microcontroller with no OS, written close to the metal (hand-optimized ARM assembly on the RP2040 path). Both do the same job, active crossovers, parametric EQ, time alignment, at a cost-of-silicon difference of roughly two orders of magnitude. Neither publishes a measured audio-quality comparison against the other or against a passive network; this spoke still has no source putting a number on what either correction layer costs or gains versus a well-designed passive crossover.
Tier and what's unverified
T3 — trade-press coverage of a hobby project, not read from the firmware source or the ASR thread. The CNX Software piece states the firmware's own specs (clock speed, bit depth, band/channel counts) rather than a vendor's marketing copy, which is why it clears T3 rather than T4, but no third party has measured the DSP's actual audio performance (THD, noise floor, filter accuracy) against the specification. The article notes Claude assisted development and that an AI code review called it "high quality for hobby software," a self-reported, unaudited claim recorded as such.
Related
camilladsp · audio-dsp · audio-dsp-hardbag-guide · crossover · group-delay · synthesis