CREPE<T>

Models & Types · AiDotNet.Audio.MusicAnalysis

CREPE (Convolutional Representation for Pitch Estimation) neural pitch detector.

Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → CREPE<T>

Implements: IPitchDetector<T>

For Beginners

CREPE listens to a single voice or instrument and tells you exactly what note is being played. It’s like a very accurate guitar tuner powered by AI. It takes small chunks of audio and outputs which pitch (frequency) is most likely present.

Usage:

How It Works

CREPE (Kim et al., 2018) is a deep convolutional network for monophonic pitch detection. It operates directly on the audio waveform (1024 samples at 16 kHz) and outputs a 360-dimensional vector representing pitch salience across 20-cent bins from C1 to B7. CREPE outperforms traditional methods (YIN, pYIN) especially in noisy conditions.

Constructors

Constructor Summary
CREPE(NeuralNetworkArchitecture<T> architecture, CREPEOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) Creates a CREPE model in native training mode.
CREPE(NeuralNetworkArchitecture<T> architecture, string modelPath, CREPEOptions options) Creates a CREPE model in native training mode.

Properties

Property Summary
double MaxPitch { get; set; }
double MinPitch { get; set; }

Methods

Method Summary
IReadOnlyList<PitchFrame<T>> ExtractDetailedPitchContour(Tensor<T> audio, int hopSizeMs)
ModelMetadata<T> GetModelMetadata()
T MidiToPitch(double midiNote)
T[] ExtractPitchContour(Tensor<T> audio, int hopSizeMs)
ValueTuple<T, T>? DetectPitchWithConfidence(Tensor<T> audioFrame)
ValueTuple<bool, T> DetectPitch(Tensor<T> audioFrame)
double GetCentsDeviation(T pitchHz)
double PitchToMidi(T pitchHz)
string PitchToNoteName(T pitchHz)