GitHub - foozmeat/NVDSP: iOS/OSX DSP for audio (Novocaine)

High-performance DSP for audio on iOS and OSX with Novocaine.

While Novocaine allows you to easily read/play audio, it is still quite hard to apply filters to the audio. This (Objective-C++) class will allow you to apply all sorts of filters (high-pass, band-pass, peaking EQ, shelving EQ etc.) in just a few lines of code.

NVDSP comes with a wide variety of audio filters:

All Pass Filter (NVAllpassFilter)
Band Pass Filter, 0dB gain (NVBandpassFilter)
Band Pass Filter, Q gain (NVBandpassQPeakGainFilter)
High Pass Filter (NVHighpassFilter)
High Shelving Filter (NVHighShelvingFilter)
Low Shelving Filter (NVLowShelvingFilter)
Low Pass Filter (NVLowPassFilter)
Notch Filter (NVNotchFilter)
Peaking EQ Filter (NVPeakingEQFilter)

Quick introduction/example (highpass filter)

// ... import Novocaine and audioFilerReader
#import "NVDSP/NVDSP.h"
#import "NVDSP/Filters/NVHighpassFilter.h"

// init Novocaine audioManager
audioManager = [Novocaine audioManager];
float samplingRate = audioManager.samplingRate;

// init fileReader which we will later fetch audio from
NSURL *inputFileURL = [[NSBundle mainBundle] URLForResource:@"Trentemoller-Miss-You" withExtension:@"mp3"];

fileReader = [[AudioFileReader alloc] 
                  initWithAudioFileURL:inputFileURL 
                  samplingRate:audioManager.samplingRate
                  numChannels:audioManager.numOutputChannels];

// setup Highpass filter
NVHighpassFilter *HPF;
HPF = [[NVHighpassFilter alloc] initWithSamplingRate:samplingRate];

HPF.cornerFrequency = 2000.0f;
HPF.Q = 0.5f;

// setup audio output block
[fileReader play];
[audioManager setOutputBlock:^(float *outData, UInt32 numFrames, UInt32 numChannels) {
    [fileReader retrieveFreshAudio:outData numFrames:numFrames numChannels:numChannels];
    
    [HPF filterData:outData numFrames:numFrames numChannels:numChannels];
}];

More examples

Peaking EQ filter

NVPeakingEQFilter *PEF = [[NVPeakingEQFilter alloc] initWithSamplingRate:audioManager.samplingRate];
PEF.centerFrequency = 1000.0f;
PEF.Q = 3.0f;
PEF.G = 20.0f;
[PEF filterData:data numFrames:numFrames numChannels:numChannels];

Lowpass filter

// import Novocaine.h and NVDSP.h
#import "NVDSP/Filter/NVLowpassFilter.h"
NVLowpassFilter *LPF = [[NVLowpassFilter alloc] initWithSamplingRate:audioManager.samplingRate];
LPF.cornerFrequency = 800.0f;
LPF.Q = 0.8f;
[LPF filterData:data numFrames:numFrames numChannels:numChannels];

Notch filter

// import Novocaine.h and NVDSP.h
#import "NVDSP/Filter/NVNotchFilter.h"
NVNotchFilter *NF = [[NVNotchFilter alloc] initWithSamplingRate:audioManager.samplingRate];
NF.centerFrequency = 3000.0f;
NF.Q = 0.8f;
[NF filterData:data numFrames:numFrames numChannels:numChannels];

Bandpass filter

There are two types of bandpass filters:

* 0 dB gain bandpass filter (NVBandpassFilter.h)
* Peak gain Q bandpass filter (NVBandpassQPeakGainFilter.h)

// import Novocaine.h and NVDSP.h
#import "NVDSP/Filter/NVBandpassFilter.h"
NVBandpassFilter *BPF = [[NVBandpassFilter alloc] initWithSamplingRate:audioManager.samplingRate];
BPF.centerFrequency = 2500.0f;
BPF.Q = 0.9f;
[BPF filterData:data numFrames:numFrames numChannels:numChannels];

Measure dB level (ranging from -51.0f to 0.0f)

// import Novocaine.h and NVDSP.h
#import "NVDSP/Utilities/NVSoundLevelMeter.h"
NVSoundLevelMeter *SLM = [[NVSoundLevelMeter alloc] init];
float dB = [SLM getdBLevel:outData numFrames:numFrames numChannels:numChannels];
NSLog(@"dB level: %f", dB);
// NSLogging in an output loop will most likely cause hickups/clicky noises, but it does log the dB level!
// To get a proper dB value, you have to call the getdBLevel method a few times (it has memory of previous values)
// You call this inside the input or outputBlock: [audioManager setOutputBlock:^...

Applying overall gain.

All sample values (typically -1.0f .. 1.0f when not clipping) are multiplied by the gain value.

// import Novocaine.h and NVDSP.h
NVDSP *generalDSP = [[NVDSP alloc] init];
[generalDSP applyGain:outData length:numFrames*numChannels gain:0.8];

Clipping

Multiple peaking EQs with high gains can cause clipping. Clipping is basically sample data that exceeds the maximum or minimum value of 1.0f or -1.0f respectively. Clipping will cause really loud and dirty noises, like a bad overdrive effect. You can use the method counterClipping to prevent clipping (it will reduce the sound level).

// import Novocaine.h and NVDSP.h
#import "NVDSP/Utilities/NVClippingDetection.h"
NVClippingDetection *CDT = [[NVClippingDetection alloc] init];
// ... possible clipped outData ...//
[CDT counterClipping:outData numFrames:numFrames numChannels:numChannels];
// ... outData is now safe ...//

// or get the amount of clipped samples:
 - (float) getClippedSamples:(float *)data numFrames:(UInt32)numFrames numChannels:(UInt32)numChannels;
// or get the percentage of clipped samples:
 - (float) getClippedPercentage:(float*)data numFrames:(UInt32)numFrames numChannels:(UInt32)numChannels;
// or get the maximum value of a clipped sample that was found
 - (float) getClippingSample:(float *)data numFrames:(UInt32)numFrames numChannels:(UInt32)numChannels;

Example project

See /Examples/NVDSPExample for a simple iOS XCodeProject example.

A thing to note:

The NVDSP class is written in C++, so the classes that use it will have to be Objective-C++. Change all the files that use NVDSP from MyClass.m to MyClass.mm.

TODO

To complete NVDSP a Spectrum Analyser would be nice. Currently it only does filtering and basic dB calculations.

Thanks to

Alex Wiltschko - Creator of Novocaine

Yasoshima - Writer of this article, revealing how vDSP_deq22 works. (and google translate, I don't speak Japanese)

hrnt - Helpful on IRC #iphonedev (freenode)