Reply to thread

Message: [QUOTE="Bombastus, post: 92293, member: 2178"] blepfx describes Spectra as both a spectral morphing engine and a spectral vocoder, but its four algorithms do not share one architecture. The overlap is real, yet the labels describe different levels of the same job. Spectra splits the main and sidechain signals into a broad spectral envelope and finer harmonic detail, then lets you recombine those parts. A conventional musical vocoder also transfers changing spectral shape between signals, but it usually does so through a bank of analysis and synthesis filters. Calling every form of envelope transfer a vocoder hides the interesting part. Spectra can estimate and apply envelopes with FFT, Hilbert, LPC, or IIR processing, and only the IIR option uses the narrow bandpass filter-bank method associated with a classic channel vocoder. [HEADING=2]Spectral envelopes are the shared core[/HEADING] A spectral envelope is the broad contour sitting over all the individual harmonics and partials in a sound. Speech makes the idea easy to hear because vowels depend heavily on resonant peaks, while the pitch can move without those resonances needing to follow note for note. Classic vocoding measures how energy changes inside several frequency bands on the modulator, then uses those changing levels to shape corresponding bands of a carrier. More bands can describe the envelope at finer frequency resolution, while fewer bands produce a coarser representation with fewer moving pieces. Cross-synthesis can reach a similar destination without relying on the same bank of filters. An FFT-based system can calculate a much more detailed spectrum, estimate its envelope, flatten the carrier, and impose another signal's envelope on it. Spectra's Linear and Hilbert modes live closer to this frequency-domain approach than to old-school channel vocoding. This distinction matters because [B][URL='https://goldmidi.com/community/threads/blepfx-has-introduced-spectra-for-spectral-morphing.77650/']Spectra's separate envelope and carrier morphing controls[/URL][/B] do more than swap a traditional modulator and carrier. You can move between the main and sidechain envelopes independently from the whitened carrier signals, so the two source roles are not locked together. [HEADING=2]IIR brings Spectra closest to a classic vocoder[/HEADING] Spectra's IIR mode uses narrow bandpass filters to extract and apply envelopes, which is the most recognizably vocoder-like architecture in the plugin. It also runs at zero latency according to blepfx, making it structurally different from the FFT-based modes before you even compare the sound. Resolution takes on a different meaning here. In Linear, Hilbert, and LPC modes, it controls FFT size, trading time resolution and latency against frequency resolution. In IIR mode, it changes the number of filter bands instead, with finer frequency resolution costing more CPU. Smooth also behaves differently in IIR mode. blepfx limits it to logarithmic smoothing and notes that lower smoothing increases phase delay, so chasing sharper spectral detail has a timing consequence even though the mode itself reports zero processing latency. Zero latency and zero phase delay are not the same claim. A classic vocoder can sound coarse because each analysis band summarizes everything happening inside its frequency range as one moving envelope. Increase the spectral resolution and more of the source's shape can survive, but the basic filter-bank behavior remains. Spectra lets you step outside that architecture completely by changing algorithms rather than only adding more bands. [HEADING=2]Cross-synthesis depends on what survives whitening[/HEADING] The less obvious part of envelope transfer is what happens to the carrier before the new envelope goes on. If its own spectral shape remains dominant, the incoming envelope has to compete with resonances already baked into the source, and the hybrid can sound muddy or stubbornly close to the original carrier. Spectra addresses this directly with whitening. Its carrier morphing stage crossfades between whitened versions of the two inputs, meaning their original envelopes are removed before the carrier blend happens. The Whiten control then limits upward expansion, which helps keep noise from being exaggerated while that flattening takes place. This idea has a long technical history. [B][URL='https://articles.ircam.fr/textes/Moorer78a/']Early LPC cross-synthesis work[/URL][/B] describes whitening a musical excitation before applying a speech-derived filter because a spectrally uneven source can obscure the transferred resonances. A richer, flatter excitation gives the new envelope more usable energy to shape. Source choice therefore matters for a deeper reason than one signal being "the vocal" and the other being "the synth." A sparse carrier with large spectral holes cannot suddenly provide strong material everywhere the borrowed envelope asks for it, while a dense carrier gives the transfer more to work with. Spectra's independent carrier and envelope morphing makes this especially useful because either input can contribute differently to the final structure. A filter-bank vocoder, FFT cross-synthesizer, and LPC system can all produce the familiar sensation of one sound speaking through another. The audible family resemblance comes from transferring spectral shape, while the machinery underneath decides how much detail survives, how timing behaves, and how strongly each source remains recognizable. [/QUOTE]

Name