Java implementation of a combination of a lightweight DSP library and an ANN for the purposes of timbre classification. ANN is very customisable, including modular layers, multiple selection functions, confusion matrices, non-binary matthews coefficient calculation, pattern shuffling, pattern selection, batch runs and graphs.
DSP library currently consists of:
Wave read/write: Read a Wave file and obtain Header info, signal info and meta data. This also allows writing of signal data and metadata into a Wave file. Can read and write up to 32bit mono and stereo files.
Signal: The basic object for signal manipulation. Multi-channel 64 bit floating point precision. Also includes methods for graph generation and adherence to fixed-point amplitude constrains.
FFT: Static fourier transform and inverse fourier transform methods. Also contains methods for easily converting a signal into an FFT signal, in addition to graphing features.
FrameFFT: Contains methods for allowing easy conversion from a signal to a windowed (or framed) FFT signal
DCT: static DCT methods and methods for easy conversion from a signal to a DCT signal
Fundamental: detection of the fundamental of a sound. (Simple algorithm but works quite well).
Pitch Shift
Filter: Implementation of 2nd order filters in the forms of: High Pass: Low Pass: Peak: Band Pass: Notch;
Mixing and gain: M/S Encoding/Decoding Summing Gain adjustment by Peak and RMS Bit Rate Conversion
Audio Generation: Generate Sine, Saw, Square, Triangle waves Generate white noise, pink noise, tape hiss
Developed for Java 1.6
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent