posted Oct 5, 2020, 8:15 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Oct 17, 2020, 9:52 PM
]
- A Comprehensive Guide to Machine Learning,
By Soroush Nasiriany, Garrett Thomas, William Wang, Alex Yang.
Department of Electrical Engineering and Computer Sciences, University of California, Berkeley. Dated June 24, 2019.
https://t.co/iZsREscT9v - The Math of Machine Learning - Berkeley University Textbook
By Garret Thomas.
needed for an introductory class in machine learning, which at UC Berkeley is known as CS 189/289A.
https://www.datasciencecentral.com/profiles/blogs/tutorial-the-math-of-machine-learning-berkeley-university - Foundations of Data Science
By Avrim Blum, John Hopcroft, and Ravindran Kannan (2018).
from Microsoft Research Lab
https://www.cs.cornell.edu/jeh/book.pdf - Machine Learning from Scratch
By Danny Friedman
author is a stats major at Harvard
https://dafriedman97.github.io/mlbook/content/introduction.html
More books here
|
posted Sep 27, 2020, 4:03 PM by MUHAMMAD MUN`IM AHMAD ZABIDI
http://viz.smultron.org/r/spectrograms/#more-94
|
posted Sep 22, 2020, 4:43 PM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 22, 2020, 5:12 PM
]
Original paper:
Results using CIFAR-10 below.
CIFAR-10 CNN as implemented by CMSIS-NN.
Performance of CMSIS-NN over baseline CNN written using arm_conv in CMSIS-DSP.
Platform: NUCLEO-F746ZG mbed board with an Arm Cortex-M7 core running at 216 MHz. Note: using state-of-art DNN, researchers has achieved ≥93% of accuracy on CIFAR-10.
Performance on STM32 compared with other SoCs, as reported by AiOTA Labs.
The players: - STM32F7 from STmicro. Up to 216 MHz.
- GAP8 from Greenwaves with CNN benchmarks. Max freq 175 MHz "Cluster", 250 MHz "Fabric Controller". RISC-V core.
- i.Mx 6ULL from NXP. This guy goes up to 900 MHz.
Looks like the GAP8 is many times more power-efficient than Cortex. At 10 FPS, the GAP8 needs 3.7 mW versus 60 mW on STM32. But it comes at a price. Very heavy price. The cost of an Arduino-compatible GAP8 board is 100,00€. And it's not running CMSIS-NN either. CMSIS-NN is for ARM processors only.
Back to CMSIS-NN.
|
posted Sep 22, 2020, 5:14 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 22, 2020, 5:19 PM
]
posted Sep 20, 2020, 3:33 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
posted Sep 20, 2020, 3:29 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 20, 2020, 3:31 AM
]
posted Sep 20, 2020, 3:08 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 20, 2020, 3:30 AM
]
posted Sep 19, 2020, 10:08 PM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 27, 2020, 11:20 PM
]
We're going to make a dataset!
- Refer to (https://www.mygardenbirdwatch.com/?cur=page/page&id=4&title=BIRD_ID) to get the 28 most common Malaysian birds.
- Remove birds not found in Sabah/Sarawak. (3 removed)
- Pick top 10 from the list in 2019. Refer to (https://www.mygardenbirdwatch.com/?cur=bird/result&date=2010). Three birds are not common in Sabah/Sarawak (starred).
4. Pick the remaining form list from (1). We're left with 25:
5. Remove those with data in Xeno-canto restricted or less than 1 hour of recordings. Now down to 13 species. Total duration for all 13 species is more 76 hours (76:48:25 or 276,505 seconds to be exact). At 44.1 kHz sampling rate, that's 24 GB uncompressed. Let's download all these sounds and then filter some more until we go down to 10 species.
Urban8k has 8732 labeled sound clips from 10 classes at 4 second each clip. That's 34,928 seconds. From the 276k seconds, we could end up with a smaller sample set after removing the silent passages. We can only know after the work is done.
- Salamon, Justin, Christopher Jacoby, and Juan Pablo Bello. "A dataset and taxonomy for urban sound research." Proceedings of the 22nd ACM international conference on Multimedia. 2014. Link at Researchgate.
- Morfi, Veronica, et al. "NIPS4Bplus: a richly annotated birdsong audio dataset." PeerJ Computer Science 5 (2019): e223. Link at publisher.
- Morfi, G. Automatic detection and classi cation of bird sounds in low-resource wildlife audio datasets. Diss. Queen Mary University of London, 2019. Link at Queen Mary.
- Salamon, Justin, et al. "Towards the automatic classification of avian flight calls for bioacoustic monitoring." PLoS ONE 11.11 (2016): e0166866. LInk at publisher.
|
posted Sep 8, 2020, 8:10 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
Amira Boulmaiz et al used TMS320C6713 DSK. Target is waterbirds in Algeria. Features: - Tonal region detector using sigmoid function
- MFCC post-processed by spectral subtraction
- SVM classifier
Papers: - Boulmaiz, Amira, et al. "Design and implementation of a robust acoustic recognition system for waterbird species using TMS320C6713 DSK." International Journal of Ambient Computing and Intelligence (IJACI) 8.1 (2017): 98-118. Link at publisher. $$$
Hervás et al used Zynq FPGA to detect Eurasian bittern (Botaurus stellaris). Paper title says endangered species. It is not. IUCN classification is "Least Concern". Features: - Wireless acoustic sensor network (WASN)
- MFCC in frames of 30 ms with 50% overlap. Filter bank of 48.
- Classifier is Gaussian Mixture Model (GMM). Computation cost is 225,920 FLOP per 30 ms frame.
- 49,156 clock cycles needed to compute MFCC or about 500 μs.
Papers: - Hervás, Marcos, et al. "An fpga-based wasn for remote real-time monitoring of endangered species: A case study on the birdsong recognition of botaurus stellaris." Sensors 17.6 (2017): 1331. Link at publisher. Open Access
Küc̣üktopcu et al used Tiva C microcontroller. Not much else is known openly. Keywords used at publisher site: - Spectral noise gating
- MFCC
Papers:
- Küc̣üktopcu, Okan, et al. "A real-time bird sound recognition system using a low-cost microcontroller." Applied Acoustics 148 (2019): 194-201. Link at publisher. $$$
|
posted Aug 24, 2020, 9:44 PM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Sep 11, 2020, 6:10 AM
]
Custom board with the Gecko processor from Silicon Labs. The Gecko contains the ARM Cortex-M4. The whole board is powered by 3 x AA batteries, uses an analog MEMS microphone and records up 384kHz. Open source. Created by 2 PhD students at the University of Southampton, Andrew Hill and Peter Prince together with Rogers, a professor at Oxford. Audiomoth can be ordered from Labmaker. Publications: - Andrew P. Hill, Peter Prince, Jake L. Snaddon, C. Patrick Doncaster, Alex Rogers. AudioMoth: A low-cost acoustic device for monitoring biodiversity and the environment. HardwareX.
- Hill AP, Prince P, Piña Covarrubias E, Doncaster CP, Snaddon JL, and Rogers A (2017). AudioMoth: Evaluation of a smart open acoustic device for monitoring biodiversity and the environment. Methods in Ecology and Evolution.
- Prince P, Hill A, Piña Covarrubias E, Doncaster P, Snaddon JL, Rogers A. Deploying acoustic detection algorithms on low-cost, open-source acoustic sensors for environmental monitoring. Sensors. 2019 Jan;19(3):553.
Uses Raspberry Pi A+ or B. The web page recommends Rode SmartLav+ microphone. From a PhD project by Sethi et al (2018) at Imperial College. Data is transmitted via GSM. Publications: - Sarab S. Sethi, Robert M. Ewers, Nick S. Jones, Christopher David L. Orme and Lorenzo Picinali (2018), Robust, real‐time and autonomous monitoring of ecosystems with an open, low‐cost, networked device. Methods in Ecology and Evolution.
- Sethi, S. S., Ewers, R. M., Jones, N. S., Signorelli, A., Picinali, L., & Orme, C. D. L. (2020). SAFE Acoustics: an open-source, real-time eco-acoustic monitoring network in the tropical rainforests of Borneo. Methods in Ecology and Evolution.
Press:
The Bela Mini is a device running Cortex and CMSIS-NN. It is created at Queen Mary University of London. It implements the 'bulbul' CNN model proposed by Grill and Schlüter, winner of the 2016 Bird Audio Detection Challenge. Publications: - Solomes, Alexandru-Marius and Stowell, Dan (2020), Efficient Bird Sound Detection on the Bela Embedded System, ICASSP 2020.
- T. Grill and J. Schlüter, Two convolutional neural networks for bird detection in audio signals, 2017 25th European Signal Processing Conference (EUSIPCO), pp. 1764-1768, Aug 2017.
Developed in collaboration with the Royal Society for the Protection of Birds (RSPB) and the British Trust for Ornithology (BTO). Customizable, its core has two parts. One, the PANDI control node (PCN) uses the ESP32 microcontroller. Two, the PANDI acoustic sampling node (PASN) uses the Raspberry Pi Zero. The system detects two hard to sample species: the Eurasian Bittern (Botaurus stellaris) and the Corn Crake (Crex crex). Data is transmitted using LoRa. Publications: Raspberry Pi based. Designed at the University of Stirling. Paired with the Cirrus Logic audio card. Combined with Peersonic RPA2 bat recorder, it became the AURITA. Publications: - Whytock, R. C., & Christie, J. (2017). Solo: an open source, customizable and inexpensive audio recorder for bioacoustic research. Methods in Ecology and Evolution, 8(3), 308-312.
- Beason, R. D., Riesch, R., & Koricheva, J. (2019). AURITA: an affordable, autonomous recording device for acoustic monitoring of audible and ultrasonic frequencies. Bioacoustics, 28(4), 381-396.
|
|
