Transform

[raymondxyy]

Interface for Common Transforms

---

audlib.sig.transform provides simple-to-use interfaces for common transforms used in speech processing by assembling functions in:

• stproc provides a useful pipeline for short-time processing
• fbanks holds useful filterbanks
• temporal holds temporal processing functions such as LPCs and zero-crossings
• spectral holds spectral processing functions such as DCT and cepstrum

Use functions in transform if you want to:

• Extract features with quick one-liners.
• Visualize spectrographic representation of audio signals.

Do not use functions in transform if:

• Speed is of critical importance. Functions such as mfcc will involve a huge amount of redundant computation for batch processing. Instead, construct Mel filterbanks with fbanks.MelFreq and compute MFCCs from there.

• Batch extract a large number of features. Chances are most features share a common pipeline (e.g. short-time framing). In that case, manually construct the pipeline with stproc.stana to avoid redundant computation.

Complete List of Supported Transforms

• stft - short-time Fourier transform
• istft - inverse short-time Fourier transform
• stlogm - short-time log magnitude spectra
• strcep - short-time real cepstra
• stccep - short-time complex cepstra
• stpsd - short-time power spectral density
• stcqt - short-time constant-Q transform
• melspec - Mel spectrogram
• mfcc - Mel frequency cepstral coefficients