MFCC Feature Extraction Pipeline

Before any acoustic model — hidden Markov, deep neural network or transformer — can recognize a phoneme, the raw waveform has to be turned into features it can actually use. This simulator runs that exact pipeline on a synthetic source-filter voice signal: pre-emphasis, Hamming windowing, a discrete Fourier transform power spectrum, a mel-scaled triangular filterbank, log compression, and a discrete cosine transform, live, frame by frame. The scrolling 3D terrain is the mel-spectrogram those filters produce; the coefficient bars are the resulting mel-frequency cepstral coefficients (MFCCs) — the compact feature vector real speech-recognition systems are built on. Switch vowels, change pitch, and dial in background noise to see how each reshapes the features an ASR model would actually receive.