Convolution Reverb audio plugin

Hello.

My idea is to create a simple convolution reverb audio plugin. I have chosen a nice Impulse response and I wrote a function script. It works and I would convert the function script in an audio plugin. For the moment I'm working with mono audio, but I will convert to stereo.

My only problem is that I have to specify also the Impulse Response as input in the function. I would find a way to "store" the IR (or more than one IR) inside the function in order to give only the audio and the parameters as input.

The IR is a too long vector and I can't explicitly write in the function code, I actually have it as a variable.

Have you any idea?

function y = ConvReverb(IR, Audio_Input, Input_Gain, Output_Gain, DryWet)

% Input and Output gain between 0 and 1. If you put 1 the volume is not
% affected
%
% Dry/Wet percentage between 0 and 1
%
% "length(IR)+length(Audio_Input)-1" is the conv length between x and h 

x = Input_Gain*Audio_Input;

d = DryWet; 
wet = d*x;
dry = (1-d)*x;

H = fft(IR, length(IR)+length(Audio_Input)-1);
DRY = fft(dry, length(IR)+length(Audio_Input)-1);
WET = fft(wet, length(IR)+length(Audio_Input)-1);

Y=H.*WET+DRY;

y_time = ifft(Y);

y = Output_Gain*y_time;

end

Answer 1

[Brian Hemmat]

Hi Lorenzo,

My understanding is that you are developing a function but eventually want to convert it to an audioPlugin object. In that case:

Take a look at the pattern in audiopluginexample.FastConvolver. It defines a property to hold the impulse response. In general, the plugins in the plugin gallery cover a number of common patterns:

https://www.mathworks.com/help/audio/ug/audio-plugin-example-gallery.html

Best,

Brian

Answer 2

[Lorenzo Lellini]

Hi Brian and thank you for your answer.

I have seen the FastConvolver and dsp.FrequencyDomainFIRFilter is what I need.

I made some tests with this object in a normal script and it works: I obtain a reverbered signals using an known impulse response.

When I convert the script into a simple plugin (I create a really simply plugin structure) it doesn't work. I had some tests on audioTestBench and I can't hear the reverbered sound when I change the dry/wet level. It is like the filter doesn't work. What is wrong?

Moreover, I have seen that dsp.FrequencyDomainFIRFilter requires a ROW VECTOR as input. In this way I have to load a mono Impulse Response and I have to transpose it because it is a column vector. Is there a way to fix it and use stereo IRs?

I'm sorry if they are quite silly questions but It is my first experience with MATLAB plugins and filter objects.

Thank you.

classdef ConvReverbPlugin < audioPlugin
    
    properties
        % PROPERTIES THAT CAN BE MODIFIED FROM THE USER.
        
        DryWet = 0.5;
        Output_Gain = 1;
        
    end
    properties (Access = private)
        % PROPERTIES NOT VISIBLE TO THE USER.
        
        % MONO IR (it is a column vector)
        IR = audioread('IR_Dataset/IR_Reverb.wav'); % MONO IR       
        
    end
    properties (Constant)
        % USER INTERFACE SECTION.
        
        PluginInterface = audioPluginInterface(...
            audioPluginParameter('DryWet', ...
            'DisplayName','Dry/Wet', ...
            'Mapping', {'lin',0,1}), ...
            audioPluginParameter('Output_Gain', ...
            'DisplayName','Output Volume','Mapping', {'lin',0,1}))
        
    end
    methods
        function out = process(plugin, in)
            % PLUGIN PROCESSING
            
            d = plugin.DryWet; 
            wet = d*in;
            dry = (1-d)*in;
            
            % I have to use the transposition ' on plugin.IR because dsp.FrequencyDomainFIRFilter
            % takes ONLY row vectors as input.
            filter = dsp.FrequencyDomainFIRFilter('Numerator', plugin.IR', ...
                'PartitionForReducedLatency', true, 'PartitionLength', 10240);
     
            y=filter(wet);
            
            out = (dry + y) * plugin.Output_Gain;

        end  
    end
end
        

[Brian Hemmat]

Hi Lorenzo,

Regarding why you're not hearing anything: You are creating the filter and using it in the process method. This means every time its called, the filter has no prior state. Instead, you need to create the filter during construction. Here is your code updated:

classdef ConvReverbPlugin < audioPlugin
    
    properties
        % PROPERTIES THAT CAN BE MODIFIED FROM THE USER.
        
        DryWet = 0.5;
        Output_Gain = 1;
        
    end
    properties (Access = private)
        % PROPERTIES NOT VISIBLE TO THE USER.
        
        % MONO IR (it is a column vector)
        IR = audioread('ChurchImpulseResponse-16-44p1-mono-5secs.wav').'; % <<<< TRANSPOSE TO ROW VECTOR      
        Filter % <<<< CREATE PROPERTY TO HOLD FILTER.
    end
    properties (Constant)
        % USER INTERFACE SECTION.
        
        PluginInterface = audioPluginInterface(...
            audioPluginParameter('DryWet', ...
            'DisplayName','Dry/Wet', ...
            'Mapping', {'lin',0,1}), ...
            audioPluginParameter('Output_Gain', ...
            'DisplayName','Output Volume','Mapping', {'lin',0,1}))
        
    end
    methods
        function plugin = ConvReverbPlugin()
            % CONSTRUCTOR - Create the filter once.
            plugin.Filter = dsp.FrequencyDomainFIRFilter('Numerator', plugin.IR, ... %<<<
                'PartitionForReducedLatency', true, 'PartitionLength', 10240); %<<<
        end
        function out = process(plugin, in)
            % PLUGIN PROCESSING
            
            d = plugin.DryWet; 
            wet = d*in;
            dry = (1-d)*in;
            
     
            y = plugin.Filter(wet); % <<< USE THE FILTER
            
            out = (dry + y) * plugin.Output_Gain;
        end  
    end
end

Regarding row vector impulse response: Just convert the impulse response to a row vector when its defined in the property. The example code above does this.

Regarding stereo convolution: Do you mean you want two separate impulse responses for the two channels? If so, create two filters and filter each channel separately. If both channels use the same impulse response, then the example code works for that. It will filter the channels independently.

Best,

Brian

[Lorenzo Lellini]

Thank you for your answer, it is now clear!