v3.7.0¶

v3.7.0 is a big release. It adds the audio module (Portenta H7 PDM microphone) with a new non-blocking callback API, the micro_speech TensorFlow speech-recognition module, a brand-new parallel-RGB LCD controller (with HDMI output and FT5X06 touch), an image.flush() method with pixel-format constants, and a rewritten image.draw_image() scaling pipeline. Several legacy helper libraries and image methods were removed and the yuv_to_* math changed — read the breaking changes below.

Highlights¶

audio module — Portenta H7 onboard PDM microphone capture with a non-blocking start_streaming(callback) API.
micro_speech — a TensorFlow micro-speech recognition module.
New LCD controller — parallel RGB display support, HDMI output (TFP410), and FT5X06 touch.
image.draw_image() — rewritten with a full scaling/alpha/palette pipeline (x_scale / y_scale / hint / color_palette …).
image.flush() plus pixel-format constants (image.BINARY / GRAYSCALE / RGB565 / YUV422 / BAYER / JPEG).
Breaking: legacy math helper libraries and several image methods were removed, draw_image() / LCD APIs were reworked, and yuv_to_* no longer subtracts 128 — see the breaking changes.

New features¶

audio — a new Portenta H7 audio module for PDM-microphone capture: audio.init(), non-blocking audio.start_streaming(callback), and audio.stop_streaming() (frequency is specified in Hz).
micro_speech — a new module with a MicroSpeech class, audio_callback(), and micro_speech() for TensorFlow speech recognition; listen() uses a running-window average and accepts a filter keyword to restrict results to a label list.
LCD — a new LCD controller with parallel RGB display support, LCD_NONE / LCD_SHIELD / LCD_DISPLAY types, many framesize constants, triple_buffer / framesize / refresh / bgr / deinit, an expanded display, HDMI output via the TFP410, and FT5X06 touch-screen support.
image.flush() — added an image.flush() method and pixel-format constants (image.BINARY / GRAYSCALE / RGB565 / YUV422 / BAYER / JPEG).
image.draw_image() — rewritten with a new scaling pipeline: x_scale / y_scale / x_size / y_size, rgb_channel, alpha, color_palette, alpha_palette, and hint, plus image.AREA / BILINEAR / BICUBIC / CENTER / EXTRACT_RGB_CHANNEL_FIRST / APPLY_COLOR_PALETTE_FIRST constants and new alpha-blending / color-table / scaling examples.
ImageReader.next_frame() gained a pause keyword to disable the real-time playback delay.
Added Portenta audio examples (audio_fft.py, micro_speech.py) and an ExtInt wake-from-stop example.

Other changes and improvements¶

The built-in examples were reorganized under scripts/examples/Arduino/; the framebuffer now updates instantly after loading/creating images with copy_to_fb=True (no manual flush needed); the Portenta heap was reduced slightly to fit more static driver state.

Bug fixes¶

Camera and imaging:

Fixed swapped red/blue (RGB565 byte order) in captured images, DMA2D draw rounding to match other draw code, fb_alloc corruption in recursive allocations (find_blobs()), and gated top_hat() / black_hat() behind both binary and math ops being enabled.

Display and audio:

Fixed the LCD shield output and the Portenta 1-channel mono audio mode.

Portenta:

Worked around an Ethernet/SDRAM issue (drive rst_eth high) and fixed the SDRAM timing configuration.

Hardware and board support¶

Arduino Portenta H7 — onboard PDM microphone (SAI / PDM2PCM) audio capture.
LCD — new parallel-RGB display controller (LTDC enabled in the H7 build), HDMI output via the TFP410, and FT5X06 touch-screen LCD support.

Breaking API changes¶

User-visible API breaks between v3.6.9 and v3.7.0. Scope: Python C-modules in modules/ and Python libraries in scripts/libraries/.

Each change is tagged with its impact:

minor — narrow API; only affects scripts that used it.
behavior — same API, different results; re-check tuned scripts.

Changes are grouped by impact in that order. If you just want to port your code, jump to the migration checklist at the end. Each commit hash links to its diff on GitHub.

Legacy math helper libraries removed (minor)¶

The mtx, rv, ulinalg, umatrix, and vec helper libraries were removed from scripts/libraries in favor of ulab. Scripts that import any of these will fail and must be ported to ulab.

Commits: 1f7da9272

`remove_shadows()` / `chrominvar()` / `illuminvar()` removed (minor)¶

image.remove_shadows(), image.chrominvar(), and image.illuminvar() were deleted with no replacement. Scripts calling these methods must remove the calls.

Commits: 3173c2bb3

`image.draw_image()` signature reworked (minor)¶

image.draw_image() was rewritten on top of a new scaling pipeline. The old positional alpha / single scale arguments and prior positional ordering no longer apply; use the new keyword form (x_scale= / y_scale= or x_size= / y_size=, alpha=, color_palette=, hint= …).

Commits: 3439f8824

LCD module rewrite (minor)¶

The lcd module was rewritten for the new display controller. lcd.init() now takes a display type (LCD_NONE / LCD_SHIELD / LCD_DISPLAY) and new keyword arguments, and the constant / API surface was substantially reorganized. Old shield-only scripts mostly still work, but the defaults and behavior changed — review LCD scripts against the new type / framesize API.

Commits: 185538207

`yuv_to_` no longer subtracts 128 (behavior)*¶

image.yuv_to_binary() / yuv_to_grayscale() / yuv_to_rgb() / yuv_to_lab() now treat the Y tuple element as an unsigned 0–255 value instead of subtracting 128. Callers must pass Y without the previous −128 bias to get the same colors as before.

Commits: dcf141192

Migration checklist¶

For a clean port to v3.7.0 the typical work is:

Port mtx / rv / ulinalg / umatrix / vec usage to ulab (the math-library removal).
Remove image.remove_shadows() / chrominvar() / illuminvar() calls (the removed image methods).
Update image.draw_image() calls to the new keyword signature (the draw_image rework).
Review LCD scripts against the new lcd type / framesize API (the LCD rewrite).
Drop the −128 bias when passing Y to yuv_to_* (the yuv_to_* change).

All other scripts run unchanged.