acoustic-signal-processing

Voice/Music acoustic signal processing techniques with explicative GUI.

Implementations

• Hidden Markov Model with MFCC as feature vector for simple voice recognition

• Non-negative Matrix Factorization for Sound Separation

• Autocorrelation / Subharmonic Summation for pitch detection

Features

• Real-time or file input

• Display multiple charts as of entire period or specified frame length.

• Display calculation and recognition results: loudness, fundamental frequency, Japanese vowel prediction,...

Chart List

Waveform (current frame / entire period)

Waveform of entire period with marker indicating current position.

Waveform of current frame.

Spectrum

Spectrogram

Spectrogram with display of voiced / unvoiced period differentiated using zero-crossing rate and fundamental frequency.

Autocorrelation

Autocorrelation and fundamental frequency position (second peak)

Spectrum + Cepstrum (frame)

Display filter spectrum (default: lifter order 13) in comparison with original spectrum.

Recognition Result

Using saved training result to determine a / i / u / e / o period.

Libraries

• Commons Math: The Apache Commons Mathematics Library

• Commons CLI

• Utilities Library (for class use)