master class: deep learning for signals · 7 deep learning is a type of machine learning in which a...
TRANSCRIPT
![Page 1: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/1.jpg)
1© 2015 The MathWorks, Inc.
Master Class:
Deep Learning for Signals
Abhijit Bhattacharjee
Senior Application Engineer
MathWorks
![Page 2: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/2.jpg)
2
Music Genre Classification
![Page 3: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/3.jpg)
3
![Page 4: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/4.jpg)
4
![Page 5: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/5.jpg)
5
Agenda
Why deep learning?
Deep learning with signal data
(Demo) Speech Command Recognition
(Demo) LSTM Networks
Enabling Features in MATLAB
Deploying deep learning
![Page 6: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/6.jpg)
6
What is Deep Learning?
![Page 7: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/7.jpg)
7
Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data.
Deep learning is usually implemented using a neural network architecture.
![Page 8: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/8.jpg)
8
Deep Learning
What is Deep Learning?
Machine
Learning
Deep
Learning
▪ Subset of machine learning with automatic feature extraction
– Can learn features directly from data
– More Data = better model
![Page 9: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/9.jpg)
9
Types of Datasets
Tabular
Data
Time Series/
Signal
Image
Data
ML or LSTMLSTM or CNN
CNN
![Page 10: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/10.jpg)
10
Image Example: Object recognition using deep learning
Training (GPU)
Millions of images from 1000
different categories
PredictionReal-time object recognition using
a webcam connected to a laptop
![Page 11: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/11.jpg)
11
Signals Example: Analyzing signal data using deep learning
Signal Classification using LSTMs Speech Recognition using CNNs
![Page 12: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/12.jpg)
12
Deep Learning Workflow
Files
Databases
Sensors
ACCESS AND EXPLORE
DATA
DEVELOP PREDICTIVE
MODELS
Hardware-Accelerated
Training
Hyperparameter Tuning
Network Visualization
LABEL AND PREPROCESS
DATA
Data Augmentation/
Transformation
Labeling Automation
Import Reference
Models
INTEGRATE MODELS WITH
SYSTEMS
Desktop Apps
Enterprise Scale Systems
Embedded Devices and
Hardware
![Page 13: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/13.jpg)
13
How Does a Convolutional Neural Network Work?
ObjectsShapes
Edges
![Page 14: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/14.jpg)
14
Speech Command Recognition
Using Convolutional Neural Networks
No
Up
Right
Off
![Page 15: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/15.jpg)
15
Example: Speech Command Recognition Using Deep
Learning
CNN Network for Audio Classification
Stop
Up
Yes
<none>
audioDatastore
auditorySpectrogram
auditorySpectrogram
audioDeviceReader
"Up"
1. Train
2. Predict
Google speech
command dataset
![Page 16: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/16.jpg)
16
Command
Recognition Demo
![Page 17: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/17.jpg)
17
![Page 18: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/18.jpg)
18
![Page 19: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/19.jpg)
19
![Page 20: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/20.jpg)
20
![Page 21: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/21.jpg)
21
![Page 22: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/22.jpg)
22
![Page 23: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/23.jpg)
23
![Page 24: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/24.jpg)
24
![Page 25: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/25.jpg)
25
![Page 26: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/26.jpg)
26
![Page 27: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/27.jpg)
27
https://www.isca-speech.org/archive/interspeech_2015/papers/i15_1478.pdf
![Page 28: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/28.jpg)
28
![Page 29: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/29.jpg)
29
![Page 30: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/30.jpg)
30
![Page 31: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/31.jpg)
31
![Page 32: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/32.jpg)
32
![Page 33: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/33.jpg)
33
![Page 34: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/34.jpg)
34
![Page 35: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/35.jpg)
35
![Page 36: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/36.jpg)
36
MATLAB Based Algorithm Wins the 2017 PhysioNet/CinC
Challenge to Automatically Detect Atrial Fibrillation
Challenge
Design an algorithm that uses machine learning to detect
atrial fibrillation and other abnormal heart rhythms in
noisy, single-lead ECG recordings
SolutionUse MATLAB to analyze ECG data, extract features using
signal processing and wavelet techniques, and evaluate
different machine learning algorithms to train and
implement a best-in-class classifier to detect AF
Results
▪ First place in PhysioNet/CinC Challenge achieved
▪ ECG data visualized in multiple domains
▪ Feature extraction accelerated with parallel
processing
“I don’t think MATLAB has any strong competitors for
signal processing and wavelet analysis. When you add in
its statistics and machine learning capabilities, it’s easy to
see why nonprogrammers enjoy using MATLAB,
particularly for projects that require combining all these
methods.”
- Ali Bahrami Rad, Aalto University
Link to user story
Block diagram for Black Swan’s atrial fibrillation detection
algorithm.
![Page 37: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/37.jpg)
37
Agenda
Why deep learning?
Deep learning with signal data
(Demo) Speech Command Recognition
(Demo) LSTM Networks
Enabling Features in MATLAB
Deploying deep learning
![Page 38: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/38.jpg)
38
I was born in France. I speak __________ ?
![Page 39: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/39.jpg)
39
Recurrent Neural Networks
▪ Take previous data into account when making new predictions
▪ Signals, text, time series
![Page 40: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/40.jpg)
40
I was born in France…
[2000 words]
… I speak __________ ?
![Page 41: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/41.jpg)
41
Long Short Term Memory Networks
▪ RNN that carries a memory cell throughout the process
▪ Sequence Problems
c0 C1 Ct
![Page 42: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/42.jpg)
42
LSTM
Demo
![Page 43: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/43.jpg)
43
Time Series Classification (Human Activity Recognition)
Long short-term memory networks
▪ Dataset is accelerometer and gyroscope
signals captured with a smartphone
▪ Data is a collection of time series with 9
channels
Deep
Learning
sequence to one
![Page 44: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/44.jpg)
44
![Page 45: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/45.jpg)
45
Agenda
Why deep learning?
Deep learning with signal data
(Demo) Speech Command Recognition
(Demo) LSTM Networks
Enabling Features in MATLAB
Deploying deep learning
![Page 46: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/46.jpg)
46
Deep Learning on CPU, GPU, Multi-GPU and Clusters
Single CPU
Single CPUSingle GPU
HOW TO TARGET?
Single CPU, Multiple GPUs
On-prem server with GPUs
Cloud GPUs(AWS)
![Page 47: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/47.jpg)
47
Audio Datastore
▪ Programmatic interface to large
collections of audio files
▪ (Optional) auto-generation of
labels from folder names
▪ Label-based indexing and
partitioning
▪ Random file sampling
▪ Automatic sequential file reading
▪ Parallel access for pre-
processing, feature extraction, or
data augmentation
ads =
Datastore with properties:
Files: {
' ...\Temp\speech_commands_v0.01\_background_noise_
' ...\Local\Temp\speech_commands_v0.01\_background_noise_
' ...\Local\Temp\speech_commands_v0.01\_background_noise_
... and 64724 more
}
Labels: [_background_noise_; _background_noise_; _background_noise_ ... and 64724 more categorical]
ReadMethod: 'File'
OutputDataType: 'double'
ads = audioDatastore('.\Dataset', ...
'IncludeSubfolders',true, ...
'FileExtensions','.wav', ...
'LabelSource','foldernames')
ads = getSubsetDatastore(ads,isCommand|isUnknown);
countEachLabel(ads)
ans =
11×2 table
Label Count
_______ _____
down 2359
go 2372
left 2353
no 2375
off 2357
on 2367
right 2367
stop 2380
unknown 4143
up 2375
yes 2377
[adsTrain,adsValidation,adsTest] = splitData(ads,datafolder);
![Page 48: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/48.jpg)
48
“I love to label and
preprocess my data”
~ Said no engineer, ever.
![Page 49: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/49.jpg)
49
Audio Labeler
▪ Work on collections of
recordings or record new audio
directly within the app
▪ Navigate dataset and playback
interactively
▪ Define and apply labels to
– Entire files
– Regions within files
▪ Import and export audio folders,
label definitions and datastores
![Page 50: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/50.jpg)
50
![Page 51: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/51.jpg)
51
Signal Analyzer App
• Navigate through signals
• Extract regions of interest
• Generate MATLAB scripts
Analyze multiple signals in time, frequency and time-frequency domains
![Page 52: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/52.jpg)
52
Feature Extraction: Time-Frequency Analysis
Reassigned spectrogram
and synchrosqueezing
Persistent spectrum Wavelet scalogram Constant Q transform
Empirical mode decomposition Kurtogram Instantaneous frequency
Instantaneous BWHilbert-Huang transform
![Page 53: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/53.jpg)
53
Enabling Features Summary
Train Predict
Ground
Truth
Labeling
Read &
PartitionRecord &
store
Process
&
augment
Extract
Features
(For Audio Applications) Audio System Toolbox
(For Signal Applications) Signal Processing Toolbox
Neural Network TB
Stats & ML TB
ads =
Datastore with properties:
Files: {' ...\Temp\speech_commands_v0.01\_background_noise_\doing_the_dishes.wav';' ...\Local\Temp\speech_commands_v0.01\_background_noise_\dude_miaowing.wav';' ...\Local\Temp\speech_commands_v0.01\_background_noise_\exercise_bike.wav'... and 64724 more
}Labels: [_background_noise_; _background_noise_; _background_noise_ ... and 64724 more
categorical]ReadMethod: 'File'
OutputDataType: 'double'
Pro
toty
pe!
![Page 54: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/54.jpg)
54
Agenda
Why deep learning?
Deep learning with signal data
(Demo) Speech Command Recognition
(Demo) LSTM Networks
Enabling Features in MATLAB
Deploying deep learning
![Page 55: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/55.jpg)
55
Deploying Deep Learning Models for Inference
Code
Generation
CNN
NVIDIA
TensorRT &
cuDNN
Libraries
ARM
Compute
Library
Intel
MKL-DNN
Library
![Page 56: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/56.jpg)
56
GPU Coder Fills a Gap in the Deep Learning Solution
Access Data Preprocess Select Network Train
Image Acq.Image
Processing
Computer
Vision
Neural
NetworkPCT
GPU
Coder
Deploy
InferenceTraining
![Page 57: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/57.jpg)
57
With GPU Coder, MATLAB is fast
Intel® Xeon® CPU 3.6 GHz - NVIDIA libraries: CUDA9 - cuDNN 7 - Frameworks: TensorFlow 1.8.0, MXNet 1.2.1, PyTorch 0.3.1
GPU Coder is faster
than TensorFlow,
MXNet and Pytorch
TensorFlow
MXNet
GPU Coder
PyTorch
![Page 58: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/58.jpg)
58
MathWorks can help you do Deep Learning
▪ Guided evaluations with a MathWorks deep learning engineer
▪ Proof-of-concept projects
▪ Deep learning hands-on workshop
▪ Seminars and technical deep dives
▪ Deep learning onramp course
▪ Consulting services
▪ Training courses
▪ Technical support
▪ Advanced customer support
▪ Installation, enterprise, and cloud deployment
▪ MATLAB for Deep Learning
Free resources More options
![Page 59: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/59.jpg)
59
Thank you!
![Page 60: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/60.jpg)
60
Extra Slides
![Page 61: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/61.jpg)
61
Machine learning vs. deep learning
Deep learning performs end-to-end learning by learning features, representations and tasks directly
from images, text, sound, and signals
Machine Learning
Deep Learning
Machine
Learning
Deep
Learning
Subset of machine
learning with automatic
feature extraction
![Page 62: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/62.jpg)
62
Feature Extraction: Spectral Analysis
Welch periodogram BW measurements Lomb periodogram
Octave spectrum Harmonic distortion Spectral statistics
![Page 63: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/63.jpg)
63
Features Extraction and Signal Segmentation
For Machine and Deep Learning
▪ Mel Frequency Cepstral Coefficients (MFCC)
▪ Pitch
▪ Voice Activity Detection
![Page 64: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/64.jpg)
64
Mel Frequency Cepstral Coefficients (MFCC)
▪ Variation of "perceptually-adjusted" spectrum content over time
▪ Most common voice and speech features for Machine Learning applications
▪ Option of:
– mfcc function,
– cepstralFeatureExtractor System object
– Cepstral Feature Extractor block (Simulink)
▪ Returns:
– MFCC
– "Delta" (= mfcc[n]-mfcc[n-1])
– "Delta delta" (= delta[n]-delta[n-1])
![Page 65: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/65.jpg)
65
Pitch Extraction
▪ Estimate audio pitch over time
▪ Popular feature for machine learning in
voice and speech processing
▪ Choice of 5 different popular algorithms
– 'NCF' – Normalized Correlation Function
– 'PEF' – Pitch Estimation Filter
– 'CEP' – Cepstrum Pitch Determination
– 'LHS' – Log-Harmonic Summation
– 'SRH' – Summation of Residual Harmonics
[ft, idx] = pitch(audioIn, fs);
![Page 66: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/66.jpg)
66
Voice Activity Detection (VAD)
▪ Standard algorithm to segment voice and speech signals
▪ voiceActivityDetector System object for MATLAB
▪ Voice Activity Detector block for Simulink
>> vad = voiceActivityDetector
vad =
voiceActivityDetector with properties:
InputDomain: 'Time'Window: 'Hann'
FFTLength: []SilenceToSpeechProbability: 0.2000SpeechToSilenceProbability: 0.1000
![Page 67: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/67.jpg)
67
TensorRT speeds up inference for TensorFlow and GPU Coder
Images / S
ec
TensorFlow
GPU Coder
cuDNN TensorRTIntel® Xeon® CPU 3.6 GHz - NVIDIA libraries: CUDA9 - cuDNN 7 - Frameworks: TensorFlow 1.6.0, MXNet 1.1.0, MATLAB 18a
![Page 68: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/68.jpg)
68
GPU Coder with TensorRT provides best performance across
all batch sizesIm
ages / S
ec
Batch Size
GPU Coder
GPU Coder/TensorRT
TensorFlow
TensorFlow/TensorRT
MXNet
![Page 69: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/69.jpg)
69
Embedded: GPU Coder also fastIm
ages / S
ec
NVIDIA libraries: CUDA9 - cuDNN 7 – Jetpack 3.2.1 - Frameworks: TensorFlow 1.6.0, MXNet 1.1.0, MATLAB 18a
![Page 70: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/70.jpg)
70
Example: Speech Denoising with Deep Learning
CNN & Fully Connected Networks for 2-D Regression
+
1. Train
2. Predict
![Page 71: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/71.jpg)
71
Example 3: Speech Denoising with Deep Learning
CNN & Fully Connected Networks for 2-D Regression
![Page 72: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/72.jpg)
72
Speech Denoising
Demo
![Page 73: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/73.jpg)
74
Extras
![Page 74: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/74.jpg)
75
Summary and Release Timeline
![Page 75: Master Class: Deep Learning for Signals · 7 Deep learning is a type of machine learning in which a model learns to perform tasks directly from image, time-series or text data. Deep](https://reader030.vdocuments.site/reader030/viewer/2022041022/5ed370cbb458607d8231caca/html5/thumbnails/75.jpg)
77