Esparagus Audio Brick¶

The Esparagus Audio Brick is built around a TI TAS5825M Class-D DAC and amplifier. Like the SqueezeAMP, it needs no external DAC — connect speakers directly.

The board exists as an ESP32 revision and an ESP32-S3 revision. The audio design is the same; the pinout is not, and neither is what the two chips can do — see Default GPIO assignments and Variants.

Some bricks are fitted with a TI TAS5805M rather than the TAS5825M. It is the same family and the same driver, but it is not the same part: see TAS5805M boards for what it does and does not do here.

Features¶

• TAS5825M with on-chip DSP and a 15-band parametric EQ (25 Hz – 16 kHz)
• Hardware volume control with a configurable maximum level
• Speaker fault detection with automatic mute and recovery
• Automatic power state management (deep sleep / standby / play) driven by AirPlay session state
• 8 MB flash
• Bluetooth A2DP on the ESP32 revision only — the S3 has no Bluetooth Classic radio, so there is no -bt build for it
• W5500 SPI Ethernet with automatic WiFi failover, on both revisions

Flashing¶

# ESP32 — AirPlay + Ethernet
pio run -e esparagus-audio-brick -t upload
pio run -e esparagus-audio-brick -t uploadfs

# ESP32 — AirPlay + Bluetooth + Ethernet
pio run -e esparagus-audio-brick-bt -t upload
pio run -e esparagus-audio-brick-bt -t uploadfs

# ESP32-S3 — AirPlay + Ethernet (no Bluetooth on this chip)
pio run -e esparagus-audio-brick-s3 -t upload
pio run -e esparagus-audio-brick-s3 -t uploadfs

# Serial monitor
pio run -e esparagus-audio-brick -t monitor

# ESP32 revision
idf.py set-target esp32
idf.py -DSDKCONFIG_DEFAULTS="config/sdkconfig.defaults;config/sdkconfig.defaults.esparagus-audio-brick" build

# ESP32-S3 revision
idf.py set-target esp32s3
idf.py -DSDKCONFIG_DEFAULTS="config/sdkconfig.defaults;config/sdkconfig.defaults.esparagus-audio-brick-s3" build

idf.py -p /dev/ttyUSB0 flash

Default GPIO assignments¶

The two revisions share no pinout, so pick the column matching your board. The ESP32 column is config/sdkconfig.defaults.esparagus-audio-brick, the S3 column config/sdkconfig.defaults.esparagus-audio-brick-s3.

The Audio Brick Dual is an S3 board with a third pinout: it has two amplifiers and no FAULTZ line back to the MCU, spending GPIO 18 and 17 on the two DAC enables instead. See Esparagus Audio Brick Dual.

The build selects the TAS58xx driver automatically (CONFIG_DAC_TAS58XX). The driver auto-detects the amplifier at startup: 0x4C–0x4F for a TAS5825M, 0x2C–0x2F for a TAS5805M. It then confirms the guess against the die ID and warns if the two disagree.

TAS5805M boards¶

Some bricks carry a TAS5805M instead. The driver detects it, drives it and gives it its own init sequence, so the board plays and its volume, mute and 15-band biquad chain all work through the Equaliser page exactly as on a TAS5825M.

Two things do not carry over:

The input mixer is TAS5825M-only too. Summing to mono or feeding one channel to both outputs is done in the DSP mixer, and only the TAS5825M's mixer is implemented here, so a TAS5805M passes the stereo pair straight through and logs a warning if asked for anything else. That also means no crossover layout needing a summed or single-channel feed — the ones that route the pair as-is still work.

Equaliser¶

TAS5825M boards expose the DAC's 15 cascaded biquad sections through the device's web interface at /bq, one filter per section, per output and per amplifier. Each section can be a peaking filter, a shelf, a low or high pass in six alignments, a band pass, a notch, a phase shift, or five raw coefficients. The filter models match PurePath Console 3, and coefficients are recomputed whenever the I2S sample rate changes. The chip's two outputs are named A and B — which of them carries left, right or a sum is the routing setting, not a fixed assignment. A and B can be ganged or tuned separately. Editing only redraws the response graph: Apply sends the filters to the amplifiers so you can hear them, and Commit to flash makes them survive a reboot. Revert goes back to what is in flash. For plain tone shaping, Load 15-band EQ fills the chain with a flat graphic equaliser — one peaking section per band from 20 Hz to 16 kHz — leaving only the gains to set. It fills in the form and nothing more, so the amplifier hears it only once applied.

Crossovers are built from these same sections, so a two-way or subwoofer split is just a high pass on one amplifier and a low pass on the other. Build crossover does that for you: pick how the drivers are wired — both bands on one amplifier, one amplifier per speaker, tweeters on one amplifier and woofers on the other, or satellites plus a subwoofer — then a crossover frequency and an alignment: Linkwitz-Riley at 12 or 24 dB per octave, Butterworth at 6, 12 or 24, or Bessel at 12. It writes the filters into every output the split touches, sets ganging and input routing to match, and leaves the filters staged so nothing is heard until applied. Apply always pushes all amplifiers, so a crossover spanning both can never go live by halves. Layouts the wiring rules out are not offered — a bridged amplifier has no separate A and B to split across.

Steeper alignments cost more slots, because sections cascade. A Linkwitz-Riley is two cascaded Butterworths of half its order, so LR2 is two 1st-order Butterworths — real poles, which collapse into a single biquad at Q 0.5 — while LR4 is two Butterworth 2 sections at Q 0.707 and cannot be folded into one. That is why a 24 dB per octave split shows two Butterworth 2 sections rather than one Linkwitz-Riley 2: both would slope at 24 dB per octave, but only the Butterworth pair sits 6 dB down at the corner, which is what lets the two branches sum flat. Two Linkwitz-Riley 2 sections would be 12 dB down there and sum 6 dB short.

Each section also has an Inv box that flips its polarity. The builder sets this itself and does not offer it as a choice, because the alignment decides it: at the corner the branches sit 180° apart at 12 dB per octave and need opposite polarity to sum flat, but they are back in phase at 24, where inverting would instead dig a notch. Only one section of a branch ever carries it — polarity belongs to the chain, and since sections multiply, inverting an even number of them cancels back to none. The box is still there by hand for drivers wired out of phase. A section left on Bypass with Inv ticked is a plain polarity flip and costs nothing else.

Fitting to a measurement¶

Fit to a measurement… turns a measured response into a correction. Load a frequency response export — REW text, or any file with a frequency and a level in its first two columns — and the fitter searches for the peaking filters and shelves that flatten it, then writes them into the chain. Only the shape is fitted, never the absolute level. The response graph gains two overlays, the measurement as it was and as it would be once corrected, so the fit can be judged before anything is applied.

Three settings matter more than the rest. The fit range decides where the effort goes: a driver rolls off at its ends by more than any filter can undo, and leaving the range wide spends filters fighting that instead of correcting the band the driver covers — the page says so when the measurement is already well down at the low limit. Smoothing sets how much detail is chased, and a sixth of an octave keeps the room modes while ignoring the fine structure that moves when the microphone does. The boost and cut limits apply to the summed correction rather than to any one filter, and the result reports how much boost was used, so the same amount can come off that output's level to keep the headroom.

Keep first N sections protects the head of the chain. The fit replaces everything past that count, so on a bi-amped speaker keep the crossover, measure each way through it, and fit into what is left; the count is filled in from any low or high passes already sitting at the top of the chain. Kept sections are slots the fitter cannot have, so Filters to use caps itself at what remains — a 24 dB per octave crossover leaves 13 of the 15. Fitted filters are staged like any other edit — nothing is heard until applied and nothing survives a reboot until committed.

Each amplifier carries its own input routing: the stereo pair as-is, summed to (L+R)/2, or one channel fed to both outputs. A bridged amplifier drives a single voice coil, so it is always fed a single channel and defaults to the sum; ganging is implicit and the stereo option is not offered. A combined response graph at the top of the page plots every active output together, so a crossover spread across both amplifiers can be read as one picture.

Show what the outputs sum to adds one more trace to that graph. Outputs feeding separate drivers add as vectors and not as curves, so two branches that each look right alone can still cancel where they overlap — and the two curves look identical either way round, which is what makes it easy to miss. The sum is taken on the complex response instead: a crossover is right when it runs flat through the corner, and a dip there means the branches are fighting, so one of them needs Inv. The trace knows only the filters, never the drivers, their spacing or the room, so it shows what the crossover is aiming at rather than what a microphone would hear.

Levels are set in the Volume section. Master is the AirPlay volume and moves everything together. Below it each output has its own level and mute, applied in the DSP input mixer ahead of the filters — so they only ever attenuate and cost no filter headroom. Use them to match drivers of differing sensitivity. Levels and mutes take effect immediately and are stored in NVS, independently of the filter commit.

Full PPC3 tuning¶

The Equaliser page covers the fifteen biquads per channel, the crossover and the levels, which is the whole signal path for most builds. A tuned dump from TI PurePath Console 3 goes further: it is the complete device configuration — clocking, I2S format, the process flow select and every coefficient — so it reaches blocks the web interface does not expose. That includes flows whose coefficient map TI never published, because a dump replays TI's own register writes rather than addresses we would have to know.

At boot the driver looks for a dump on SPIFFS and, if one is present, replays it instead of the built-in init sequence. PPC3 bakes every coefficient at the rate the flow was exported for, so a 48 kHz tuning played at 44.1 kHz puts every corner about 8% low: name the file for its rate and the driver picks the one matching what it is playing.

Drop the -<rate> suffix — tas5825m_fw.bin, tas5825m_fw0.bin — for a dump that should serve every rate. Those names are the fallback, searched only once no rate-specific file matches, so a single-rate install keeps working untouched. A process flow is a TAS5825M feature, so the driver only looks for any of these on that part — see TAS5805M boards.

To install one:

1. Tune the part in PPC3 and export either the I2C log (.cfg) or the C header, at each sample rate you want covered.
2. Convert it:

   python3 components/dac_tas58xx/ppc3_convert.py my_tuning_44k1.cfg -o tas5825m_fw-44100.bin
   

   A log that drives both amplifiers carries writes for each, so pick one with --dev 98 or --dev 9a and convert it twice.
3. Copy the result to data/hf/ for a serial flash, or upload it over WiFi:

   curl -X POST "http://<device-ip>/api/fs/upload?path=/spiffs/hf/tas5825m_fw-44100.bin" \
        --data-binary @tas5825m_fw-44100.bin
   

4. Reboot.

Delete the file to go back to the built-in flow.

The biquad addresses are identical in every documented flow on both the TAS5825M and the TAS5805M, so the Equaliser page keeps working on top of a dump. A dump that selects an undocumented flow is the exception: nothing guarantees its coefficients live where the page expects them.

Variants¶

The ones marked prebuilt are in the browser installer. On the ESP32 the published binary is the Bluetooth one, so build esparagus-audio-brick yourself if you want that RAM back.

Bluetooth Classic exists only on the original ESP32, so the -bt build has no S3 equivalent; an S3 brick reaches the network over Ethernet or WiFi. The Dual has two amplifiers and a pinout of its own, and gets its own page.

Esparagus Louder¶

The Esparagus Louder is the same amplifier design on a board of its own.

pio run -e esparagus-louder-s3 -t upload
pio run -e esparagus-louder-s3 -t uploadfs

Related¶

• Bluetooth A2DP
• Ethernet (W5500)
• Build environments