Nicla Vision ML Model Firmware Issues

WatersHL · March 27, 2025, 5:23am

Hello,

I’ve been trying for quite some time to run machine learning models capable of detecting bird songs in the arduino nicla vision board as part of my masters degree.
However, I’ve been faced with many challenges. One of them uploading an already trained model. After following the library and training my model in the repository, I seem to be running into countless dead-ends on the subject of uploading and running the model.
I know that I need to upload it to firmware since the memory isn’t enough and that the process was once explained by a guide in the link “tflite-micro/tensorflow/lite/micro/examples/micro_speech/train at main · tensorflow/tflite-micro · GitHub” which unfortunately is not available any more, I can’t find anything regarding it. Could anyone help me with this? Did the link change?

Thank you for your patience in advance,
I feel like it’s not going to be the last time I write here.

iabdalkader · March 27, 2025, 8:42am

The guide you referenced is just a MicroSpeech training guide from Tensorflow, which is unrelated to this topic. To embed models into a firmware image, add them here and rebuild the firmware (or just replace micro_speech.[tflite/text]):

To use MicroSpeech models, we have a built-in app that implements the algorithm (starting audio, building & updating the spectrogram, running inference etc..) here:

github.com/openmv/openmv

scripts/libraries/ml/ml/apps.py

master


      
          from ml import Model
          from micropython import const
          from ulab import numpy as np
          
          try:
              import audio
          except (ImportError, AttributeError):
              pass
          
          
          class MicroSpeech:
              _SLICE_SIZE = const(40)
              _SLICE_COUNT = const(49)
              _SLICE_TIME_MS = const(30)
              _AUDIO_FREQUENCY = const(16000)
              _SAMPLES_PER_STEP = const(10 * (_AUDIO_FREQUENCY // 1000))  # 10ms * 16 Samples/ms
              _CATEGORY_COUNT = const(4)
              _AVERAGE_WINDOW_SAMPLES = const(1020 // _SLICE_TIME_MS)
          
              def __init__(self, preprocessor=None, micro_speech=None, labels=None, **kwargs):
                  self.preprocessor = preprocessor

This lets you pass the name of the preprocessor and microspeech models as arg if they’re different than the defaults. Note if the same hyperparemeters (slice size, window size etc) were used, this will just work right out of the box, if not, it will be hard to adapt but not impossible. Here’s an example using the default:

github.com/openmv/openmv

scripts/examples/03-Machine-Learning/00-TensorFlow/micro_speech.py

master

# This work is licensed under the MIT license.
# Copyright (c) 2024 OpenMV LLC. All rights reserved.
# https://github.com/openmv/openmv/blob/master/LICENSE
#
# The MicroSpeech module is designed for real-time audio processing and speech recognition
# on microcontroller platforms. It leverages pre-trained models for audio preprocessing and
# speech recognition, specifically optimized for detecting keywords such as "Yes" and "No".
import time
from ml.apps import MicroSpeech


def callback(label, scores):
    print(f'\nHeard: "{label}" @{time.ticks_ms()}ms Scores: {scores}')


# By default, the MicroSpeech object uses the built-in audio preprocessor (float) and the
# micro speech module for audio preprocessing and speech recognition, respectively. The
# user can override both by passing two models:
# MicroSpeech(preprocessor=ml.Model(...), micro_speech=ml.Model(...), labels=["label",...])
speech = MicroSpeech()

This file has been truncated. show original

WatersHL · April 1, 2025, 1:15pm

Hi,

Thank you for your previous response. While I appreciate the guidance on MicroSpeech, my core issue remains unresolved: I’m unable to rebuild the firmware for the Nicla Vision to embed custom ML models.

My current problem is that when following the official build instructions (Docker or Linux), the process fails with:

“make: *** No rule to make target ‘.WAIT’, needed by ‘openmv’. Stop.”

Would you happen to know any of these questions?

Is .WAIT a known issue with Nicla Vision builds? If so, is there a patch or workaround?
Are there additional dependencies not documented in the guide?

I’m happy to provide logs or test patches. This is a blocker for my thesis, and I’d love to help improve the docs if we solve it.

Thank you for your time!

kwagyeman · April 1, 2025, 5:06pm

Hi, we have an automated build system in the cloud you can follow. Here’s the script: openmv/.github/workflows/firmware.yml at master · openmv/openmv · GitHub

It uses this tool: openmv/tools/ci.sh at master · openmv/openmv · GitHub

Please follow the workflow steps; you should have no problem building the firmware.

iabdalkader · April 2, 2025, 7:30am

The docker build has been fixed.

Topic		Replies	Views
Tflite Model in OpenMV OpenMV Pure Thermal	0	134	February 15, 2024
OpenMV can't build a new firmware for Nicla Vision Arduino Boards	3	517	April 7, 2023
Nicla Vision - Micro Speech Example Code, No Response and No Recognition Arduino Boards	3	312	October 11, 2023
Microspeech Arduino Boards	2	572	January 25, 2021
Having Trouble building firmware for Nicla Vision Arduino Boards	1	301	May 25, 2023

Nicla Vision ML Model Firmware Issues

Related topics