Skip to main content
k2k audio logo k2k audio

Back to Extractors
Documentation tree

Spectral Envelope Extractor

Speech and many instrument sounds can be modeled as a "source" (the buzzing of vocal cords, the buzz of an oboe reed, the noise of a snare) passed through a…

Parameters

ParameterRangeDefault
Cepstral Order20.0 – 80.040.0
Iterations5.0 – 20.010.0

Cepstral Order — Order of the cepstral analysis used to extract the spectral envelope, 20–80. Higher orders track the envelope more precisely (more detail in the timbral profile) but can pick up pitch information as part of the envelope, defeating the purpose. 40 is a balanced default for most material; lower (25–35) for cleaner timbre/pitch separation; higher (60–80) for highly-detailed envelope tracking on relatively-fixed-pitch sources.

Iterations — Number of refinement iterations for envelope extraction, 5–20. The algorithm refines its estimate of the envelope vs. excitation split iteratively. More iterations = more accurate separation but more CPU. 10 is balanced; 15–20 for high-quality offline rendering.

Additional controls

Output — Which component to extract on the primary output:

  • Envelope — the slowly-varying spectral envelope (timbre/formant/filter shape — the “color” of the source). Extracted = envelope, rejected = excitation.
  • Excitation — the pitch/source content that sits underneath the envelope (the harmonic structure, the pitched buzz before the vocal-tract filtering). Extracted = excitation, rejected = envelope.

About Spectral Envelope Extractor

Speech and many instrument sounds can be modeled as a “source” (the buzzing of vocal cords, the buzz of an oboe reed, the noise of a snare) passed through a “filter” (the resonances of the vocal tract, the body of the oboe, the resonances of the snare drum body). This source-filter model is the foundation of vocoders, formant filtering, and a lot of speech synthesis. This node decomposes the input into those two streams: the envelope (the filter) and the excitation (the source). Use the envelope output as a vocal tract shape you can apply to other sources (cross-spectral filtering for vocoder effects); use the excitation output to study or process the underlying buzz independent of the timbre. For voice-specific formant extraction (just the resonant peaks, not the full envelope), use the Formant Extractor instead.


Generated 2026-05-05 from K2K_Dev@96730bdc by scripts/gen_lexique.py. Edit _intros/ or _overrides/, not this file.