“MATLAB in the Loop” for Audio Signal Processing

Darel A. Linebarger, Ph.D.

Senior Manager, Signal Processing and Communications
MathWorks, Inc.

© 2013 The MathWorks, Inc.

1
Introduction: Who am I and why am I here?

 Why: To demonstrate that you can use MATLAB and your laptop to
develop and test real time audio signal processing algorithms
 Who:
– I manage a development group at MathWorks focused on DSP and Communications
 Includes fixed-point modeling and deployment to C or HDL
– Audio is a focus area for DSP System Toolbox
 What:
– I am on the road to channel customer input directly into development
– I am seeking a few customers to work closely with and by helping them succeed, to
make our tools better for everyone.
– Could you be one of those customers?

2
Goals for today

 Help you use our tools better so that you are

– More productive
– More efficient
 Take your input for our product plans to help you with your workflow(s)
– What new directions should we be considering?

 Initiate contact with key people or groups to help drive this area forward.

 NOTE: Most of today’s presentation is also available as a webinar from

DSP System Toolbox product page on mathworks.com

3
Where would you like us to invest next?
How can we best help you?

 What customers are saying: We want plugin support (autogenerate for

deployment, hosting)
– Which plugin formats? (Apple, VST, etc.)
 Other possible priorities:
– Performance:
 How many biquads can we run and maintain real-time?
 Reduce latency in our processing chain?
– Asynchronous sample rate conversion
– More audio algorithms
 Codecs? Recognition? Effects for music production?
– More drivers or environments (OSC, JACK, JUCE, WASAPI, etc.)
– Your good idea goes here …
 What would you suggest?
4
Start with demos

 Live audio to scopes and file

 Simple demo:
– audioIn=dsp.AudioRecorder('SamplesPerFrame',1e5, 'NumChannels', 1)
– sound(yin,44100)
– audioFileOut=dsp.AudioFileWriter;
– step(audioFileOut,yin);
– release(audioFileOut);
 Parametric equalizer

5
I said “real time”. What did I really mean?

 A laptop does not provide a true real time environment. On the other hand,
if it can process the data fast enough and reliably enough, it might work just
fine.
– E.g. We use our PCs for voice and video (Skype) communications frequently. That’s
real time communications.
 For audio signal processing, real time is only important when either or both
input and output are live audio.
– Audio input comes from microphone, audio output goes to speakers or headphones.
 What about latency?
– Not important if either input or output are not live. E.g. consider playing recorded music.
As long as the latency is not ridiculous, users will not notice it.
– If both input and output are live, then latency must be small (< 30 ms).
– We have a shipping example in 14a demonstrating how to measure latency.
6
New scopes in DSP System Toolbox

 Visualizations
– Time Scope
– Spectrum Analyzer
– Logic Analyzer

7
How to create a streaming test bench

Spectrum Analyzer

View it

Speak it
Spectrum
Analyzer

Microphone
Hear it
Audio Input Visualize sound in real-time

Speaker

Audio Output

9
How to create test bench in MATLAB

%% Create and Initialize

SamplesPerFrame = 1024;
Fs = 44100; Initialize
Microphone=dsp.AudioRecorder('SamplesPerFrame',SamplesPerFrame);

Spectra=dsp.SpectrumAnalyzer ('SampleRate',Fs);
%% Stream processing loop
tic;

while toc < 20 Process

% Read frame from microphone
audioIn = step(Microphone); in-the-loop
% View audio spectrum
step(Spectra,audioIn);
end

%% Terminate Terminate
release(Microphone)
release(Spectra)

10
Use test bench App from in product example to create a test
bench

11
DSP System Toolbox audio related components
(supported on Apple/Windows/Linux)
– Multichannel audio I/O (Number of channels depends on hardware)
 Audio Player/Recorder - Supports multiple devices, one sound driver per MATLAB session
 Audio File Reader/Writer
 ASIO low latency driver support on Windows(R)
 Custom channel mapping
– Audio signal analysis
 Scopes: time, spectrum analyzer, array plot
 Transfer function estimator
 Measurements: Average power, PeaktoRMS ratio, mean, variance, ...
– Signal processing algorithms
 FIR, Biquad, Multirate FIR, FFT, LMS, ...
 Variable fractional delay (useful for audio beamforming)
– Connectivity
 UDP, MIDI (simultaneous support for multiple controls on multiple devices)

12
Audio I/O with MATLAB: The gear

13
Audio Hardware is Automatically Detected

 Audio device I/O components (in both MATLAB and Simulink) detect audio
devices registered with OS and dynamically populate pick lists

14
Choice of Modern File Formats Allows Interplay with other
Common Audio Players

15
Audio demos

 Live feed into scopes and file write

 Sample rate conversion
 Parametric equalizer with run time interaction (real time on laptop)
 Auto generate code for audio test bench
 Fourier
 Reverb (uses ASIO)
 Plugin with Reaper

16
Optional additional topics

 Latency – measuring, minimizing, ASIO (for low latency on Windows)

 Filter design
– Sample Rate Conversion

 Plugins – generating them from MATLAB Code

 Codecs – speech or audio

 Code generation for acceleration or deployment

17
Filter Design and Sample Rate Conversion

 filterbuilder
– Generates MATLAB code for given design
– Optionally generates HDL code
 Sample rate conversion
– Baseline: FIR Decimation, FIR Interpolation and FIR Rate Converter
– New design assist: dspdemo.SampleRateConverter (see associated demo in 14a)
 Preview of 14b sample rate converter
– Allows “tolerance” to find smaller factors if approximate rate conversion acceptable
– Can reduce number of operations and/or number of stages
 Is there interest in Asynchronous sample rate conversion?
– What would you expect for interface?
 Demo
18
 DSP System Toolbox *
Over 300 algorithms for modeling, designing, implementing and deploying dynamic
system applications
●Advanced Filter Design, Adaptive, Multistage and Multi-rate Filters ●Visualization in Time and Frequency-domain
●FFT, DCT & other Transforms ●System objects and functions in MATLAB
●Signal processing blocks for Simulink ●Stream signal Processing
●Support for Fixed-Point, C/C++ code generation and HDL ● ARM Cortex-M support for hardware prototype
Algorithm libraries in MATLAB Algorithm libraries in Simulink

*http://www.mathworks.com/
products/dsp-system/index.html

21
 THANK YOU!

© 2014 The MathWorks, Inc. MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand names may be
23
trademarks or registered trademarks of their respective holders.
Where would you like us to invest next?
How can we best help you?

 What customers are saying: We want plugin support (autogenerate for

deployment, hosting)
– Which plugin formats? (Apple, VST, etc.)
 Other possible priorities:
– Performance:
 How many biquads can we run and maintain real-time?
 Reduce latency in our processing chain?
– Asynchronous sample rate conversion
– More audio algorithms
 Codecs? Recognition? Effects for music production?
– More drivers or environments (OSC, JACK, JUCE, WASAPI, etc.)
– Your good idea goes here …
 What would you suggest?
24
Agenda

 Tunable parametric equalizer example

 Dynamic range audio expander example
How to create a streaming test bench for audio processing
1 in MATLAB

How to develop algorithms and incorporate them into the

2 test bench

How to accelerate simulation for real-time performance

3

25
Stream processing in MATLAB

 Streaming techniques* process continuous data from a captured signal or large file by
dividing it into “frames” and fully processes each frame before the next one arrives
 Memory efficient
 Streaming algorithms in DSP System Toolbox provide
 Implicit data buffering, state management and indexing
 Simulation speed-up by reducing overhead

MATLAB Memory

Stream Stream
Source Processing

*http://www.mathworks.com/discovery/stream-processing.html
26
Tunable parameter equalizer example
Tune parameters
in real-time

See it

Tune it

Audio Input Array Plot

Guitar10min.ogg Parameter
a 44.1Khz stereo Equalizer
audio Filters
Audio Output Visualize audio
Play it Custom Audio Algorithm waveforms in real-time

Create it
Speaker
Hear it
27
Part 1: Test bench and peripheral access

How to create a streaming test bench for audio processing

1 in MATLAB

How to develop algorithms and incorporate them into the

2 test bench

How to accelerate simulation for real-time performance

3

28
Part 2: Algorithms

How to create a streaming test bench for audio processing

1 in MATLAB

How to develop algorithms and incorporate them into the

2 test bench

How to accelerate simulation for real-time performance

3

29
 Example 1: Dynamic audio range expander
Audio Input Custom Audio Algorithm Audio output
x

Gain

Level Gain
LPF
measurement processor

Spectrum Analyzer Time Scope

Expand
dynamic
A range B
(dB)
(dB)

Webinar: Digital Filter Design Made Easy (https://www.mathworks.com/company/events/webinars/wbnr64770.html) 30

How to incorporate algorithm into test bench
%% Create & Initialize
SamplesPerFrame = 1024;
Fs = 44100;
Microphone = dsp.AudioRecorder('SamplesPerFrame');
MyTimeScope = dsp.TimeScope('SampleRate',Fs); initialize
h = fdesign.lowpass('fp,fst,ap,ast',4750,5250,0.5,80,Fs);
FIR =design(h,'equiripple','MinOrder','any,,’StopbandShape','flat','SystemObject', true);
z = zeros(1,Microphone.NumChannels);

%% Stream processing loop

tic;
while toc < 15
% Read frame from microphone
audioIn = step(Microphone);
Process
% Dynamic range expansion algorithm in-the-loop
[audioOut,z]=expander_vec(audioIn,FIR,z);
% View audio waveform
step(MyTimeScope,[audioIn,audioOut]);
end

%% Terminate
release(Microphone) Terminate
release(MyTimeScope)
31
Example 2: Tunable audio parametric equalizer

Tune parameter
equalizer in real-time

Tune it See it

X H
H=Y/X
Array Plot
Y Transfer Function
Play it
Estimator
Guitar file Parameter Y
Hear it
@ 44.1Khz Equalizer
stereo audio
X Filters Y
Audio Input
Speaker
Custom Audio Algorithm Audio Output
Visualize audio waveform
in real-time
Create it

32
How to incorporate algorithms into test bench
%% Create & Initialize
SamplesPerFrame = 1024;
FReader=dsp.AudioFileReader('guitar2min.ogg','SamplesPerFrame’,SamplesPerFrame)
Fs = FR.SampleRate; Initialize
TransferFuncEstimate= dsp.TransferFunctionEstimator('SampleRate',Fs,’FrequencyRange','onesided','SpectralAverages',1);
MyArrayPlot = dsp.ArrayPlot('PlotType','Line','Title','Transfer Function Estimate');
Speaker = dsp.AudioPlayer('SampleRate',Fs);
GUI = CreateParamTuningGUI(param, 'Tuning');

%% Stream processing loop

tic;
while ~isDone(FR)
audioIn = step(FReader); % Read frame from file
[pauseSim,stopSim,tunedparams] = callbacks(param);
Process
% Audio processing algorithms – custom algorithm - PEQ
audioOut = audio_algorithm_peqso(audioIn,tunedparams); In-the-loop
H = step(TransferFuncEstimate,audioIn,audioOut);

step(MyArrayPlot,20*log10(abs(H))); %View estimated transfer function

step(Speaker,audioOut); %Play resulting audio
End
%% Terminate
release(FReader)
release(TransferFuncEstimate) Terminate
release(MyArrayPlot)
release(Speaker);
close(GUI);

33
Part 3: Acceleration of simulation

How to create a streaming test bench for audio processing

1 in MATLAB

How to develop algorithms and incorporate them into the test

2 bench

How to accelerate simulation for real-time performance

3

34
Stream processing:
Data acquisition & algorithm times

Frame 1 Frame 2 Frame 3 Frame n

Data acquisition Algorithm processing time

As long as

Data acquisition + Algorithm processing <= Frame time

We have
Real-time signal processing

35
 MATLAB to C code generation* MATLAB Coder

function y = audio_algorithm_peqso(u,tunedparams)
% Copyright 2013 The MathWorks, Inc.
persistent PE1 PE2
if isempty(PE1)
PE1 = parametricEQFilter('Bandwidth',2000,…
'CenterFrequency',3000,'PeakGaindB',6.02);
PE2 = ParametricEQFilter('Bandwidth',2000,…
'CenterFrequency',1000,'PeakGaindB',-6.02);
end
[PE1,PE2] = processtunedparams(tunedparams,PE1,PE2);
v = step(PE1,u);
y = step(PE2,v);
%-------------------------------------
function [PE1,PE2] = processtunedparams(tunedparams,PE1,PE2)

if ~isnan(tunedparams.CenterFrequency)
PE1.CenterFrequency = tunedparams.CenterFrequency;
end
if ~isnan(tunedparams.Bandwidth)
PE1.Bandwidth = tunedparams.Bandwidth;
end
if ~isnan(tunedparams.Gain)
PE1.PeakGaindB = tunedparams.Gain;
end
if ~isnan(tunedparams.CenterFrequency2)
PE2.CenterFrequency = tunedparams.CenterFrequency2;
end
if ~isnan(tunedparams.Bandwidth2)
PE2.Bandwidth = tunedparams.Bandwidth2;
end
if ~isnan(tunedparams.Gain2)
PE2.PeakGaindB = tunedparams.Gain2;
end

(*) Design and Prototype Real-Time DSP Systems with MATLAB (Conference Presentation):
http://www.mathworks.com/company/events/conferences/matlab-virtual-conference/2013/proceedings/design-and-prototype-real-time-dsp-systems-with-matlab.html
36
Simulation acceleration benchmarks

2-band parametric equalizer algorithm Processing time

MATLAB code 23.37 seconds

MEX code 2.84 seconds

37
 Audio signal processing is everywhere!
Tablet/ MP3 Player Gaming System
& Smart Phone

Automotive Audio
& Navigation System

Professional
Audio & Music Medical Devices
Hearing Aids

39
Create Your Own System Objects

classdef RemoveMean < matlab.System  classdef defines a System object using

% Remove estimated running mean. matlab.System
properties
% Memory weighting  properties defines parameters
Weight = 0.8
end and states of your system

properties (DiscreteState)
Mean = 0 % Initial value  methods implements the functions specific to
end your system
methods (Access=protected)
function y = stepImpl(obj,u)  stepImpl implements the kernel
y = u - obj.Mean; of the step function
obj.Mean = u - y*obj.Weight;
end
end
 Other methods to consider:
end setupImpl, resetImpl, releaseImpl

http://www.mathworks.nl/help/dsp/basic-operations.html
40