Cannot read correct frequency for AudioBuffer

[marttini18]

Hello,

I started to learning JUCE.
As the first project I wanted to create simple guitar tuner but I face some difficulties in frequency detection.
I got into tutorials and studied some of them - especially FFT connected one.

Let me show some clue part of my code.
Here are some definitions from my juce::AudioProcessor derived class.

constexpr static auto fftOrder = 10;
constexpr static auto fftSize = 1 << fftOrder;

juce::dsp::FFT forwardFFT{ fftOrder };
std::array<float, fftSize * 2> fftData{};

double sampleRate{};

For simplicity let’s assume num of samples is same as fftSize - to avoid additional FIFO.
In my processBlock function I copy data from buffer into fftData then I run performFrequencyOnlyForwardTransform and I try to get frequency from it.

I’ve done some research and I found out I need to find peak in that spectrum and then from: Freq = (sampleRate * fftBufferIdx) / bufferSize I can get frequency.

for (int channel = 0; channel < totalNumInputChannels; ++channel)
{
    const auto* channelData = buffer.getReadPointer(channel);
    const auto numSamples = buffer.getNumSamples();

    std::fill(fftData.begin(), fftData.end(), 0.0f);
    std::copy(channelData, channelData + numSamples, fftData.begin());

    forwardFFT.performFrequencyOnlyForwardTransform(fftData.data());

    const auto maxElementIndex = std::distance(fftData.begin(), std::max_element(fftData.begin(), fftData.begin() + fftSize / 2));
    const auto freq = (sampleRate * maxElementIndex) / fftSize;
    std::cout << freq << std::endl;
}

Here is the main part of my code but I does not seem to work properly.
Any help would be appreciated.

[marttini18]

Has anyone any idea what’s wrong with my code?
Or could someone give some hint?

[nicolasdanet]

Reliable pitch tracking is not an easy task.
Do you need help about the coding part or the algorithm part?
For the algorithm part there are some threads about it already on the forum.

Pitch recognizer C++ API

Hi there, I’m looking for a straightforward pitch recognizer C++ API (like every real tuner do). Does anybody here know how to find this kind of API? What I would like to do is basically extract the fundamental frequency from a given wave sample (in real-time), then convert it to the closest pitched note, and finally into a proper NoteOn MIDI signal. I already tried to Google it, but I haven’t found anything so far. Any ideas? Thanks

Revising pYIN to use JUCE's fft

I am trying to rewrite the fastDifference function in the pYIN by Matthias Mauch to use JUCE’s fft. I am doing something wrong, but I don’t know what. Here is the original: void YinUtil::fastDifference(const double *in, double *yinBuffer) { int frameSize = 2 * m_yinBufferSize; int halfSize = m_yinBufferSize; double *audioTransformedComplex = new double[frameSize + 2]; double *audioOutReal = new double[frameSize]; double *kernel = new double[frameSize]; double *kernelTransformedComplex = n…

Working with AudioBuffer to do Pitch Detection on Low Frequencies

I am working on a AudioPlugin that does pitch detection as well as play style recognition on an electric bass. I found this thread on pitch tracking in JUCE but haven’t had much success with adamski’s module tracking low frequencies so I am looking at implementing my own or modifying what he has done, but I am not entirely clear on how JUCE AudioBuffer works in tandem with other audio programs running. For example: If I need an audio buffer or 4096 samples or possibly higher for detection of s…

[benvining]

Here is an example of a pitch detector written with Juce: Shared-code/pitch-detector.h at main · benthevining/Shared-code · GitHub

Feel free to use this if you want

[marttini18]

Thanks! With your code it seems to work properly.
I am testing it with guitar plugged-in

I didn’t expect it will be such complex issue as I am not familiar with audio engineering.
I expected to just run FFT, find a peak and calculate estimated frequency based on index of that peak and fft size. It occurs not to be so easy!

[benvining]

I’m glad to hear my pitch detection code is working for you!

yeah, it’s a pretty common misconception that FFT frequency bins are analogous to fundamental frequency. There may be more energy in a frequency bin that is not actually the fundamental - like an overtone or undertone, this would lead to an “octave error”.

What my pitch detector is doing is finding the period of the fundamental frequency by determining at what time interval the signal is most periodic – so it turns out it’s actually entirely a time domain operation, nothing here is being done in the frequency domain

[nicolasdanet]

@benvining: Is it a hand-made algorithm, or is it based on a paper somewhere on the web?

[benvining]

It’s sort of cobbled together based on various sources I’ve found… papers, forum posts, StackExchange, etc. The actual periodicity calculation is a fairly standard time domain ADSF; the real magic is in the post processing “peak picking” algorithm, to prevent octave errors - that’s definitely a bit more home brewed…

Following my own intuition and some testing results, what I ended up doing was firstly placing some constraints on the detected period – namely, assuming that the period shouldn’t halve or double between consecutive pitched frames. Then, between consecutive pitched frames, I weight each autocorrelation value proportionally to how far away that period candidate is from the last detected period – ie, in order for a new period far away from the last one to be chosen, it has to be really significantly periodic.

This algorithm could definitely be improved, I’m sure it’s a bit naive in some ways, but it seems to do pretty well with detecting pitch of a vocal input, which is my main use case ¯_(ツ)_/¯

[nicolasdanet]

Thanks for the explanations.

[benvining]

of course!