Mel-cepstral distortion (MCD) is an objective metric that approximates the perceptual spectral difference between a reference utterance and a processed or synthesized utterance by comparing their mel-frequency cepstral coefficient vectors frame by frame. For each aligned frame pair, with c_i the processed-utterance coefficients and r_i the reference coefficients (typically i = 1..L with L = 12 or 13; the 0th coefficient c_0, overall level, is excluded or weighted), MCD(frame) = 10·log10( sqrt( sum_i (c_i - r_i)^2 / sum_i r_i^2 ) ) dB, and the reported value is the average over the overlapping frames. Parameters: (1) coefficient order range L, (2) c_0 treatment (excluded or weighted), (3) frame alignment and overlap interval. Persistence: standardized objective metric for TTS and speech-enhancement evaluation in the ITU-T P.56 lineage, implemented in the HTS toolkit, WORLD vocoder utilities, and OpenSMILE, and routinely reported in dB in the ICASSP and SLT speech-synthesis literature. [formal: distortio cepstralis mel | substrate: behavior | horizon: a moment | explicit: yes | epoch: 0.01]
Accepted ontology entry
mel-cepstral-distortion
Mel-cepstral distortion (MCD) is an objective metric that approximates the perceptual spectral difference between a reference utterance and a processed or synthesized utterance by comparing their mel-frequency cepstral coefficient vectors…
Definition
Why it is in scope
Human-made objective metric for speech quality assessment, built to persist in audio engineering literature and evaluation pipelines.
Names and aliases
- mel-cepstral-distortionen · CANONICAL
Relations from this entry
- cmspxgu9f06vtjlssp8aaz07zDEPENDS_ON →
MCD is computed from mel-cepstral coefficients derived via mel-filterbank analysis of the spectrum. Remove mel-filterbank and the cepstral coefficients cannot be formed, so MCD stops operating.
- cmspywurr06ztjlssi76045rfSERVES →
Mel-cepstral distortion is a metric computed from mel-frequency cepstral coefficients to quantify spectral distortion between reference and processed speech; it is built for the sake of evaluating and improving speech recognition and synthesis systems.
- cmsr1p0gz0020kp539o6mpun7DEPENDS_ON →
MCD computation requires MFCC vectors as input; remove MFCC computation and MCD cannot operate — removal test satisfied, Law 8. Dependency is object-level, not meta-level.
- cmsps9i0v06eejlssqrjcqyviDEPENDS_ON →
Mel-cepstral-distortion quantifies perceptual difference between utterances through mel-cepstral coefficient computation — a signal-processing technique. Remove signal-processing (the mathematical analysis and transformation of audio signals) and mel-cepstral-distortion cannot be computed or used. The metric depends on signal-processing for its core computational mechanism.
- cmsqhjrqb000lti676a6qcrwrDEPENDS_ON →
Mel-cepstral distortion is a distance metric computed from mel-frequency cepstral coefficients — it quantifies spectral distortion between reference and processed speech by comparing coefficient vectors. The removal test passes: without cepstral coefficients as the computational substrate, the distortion metric cannot be computed or operate. Direction: mel-cepstral-distortion depends on cepstral-coefficients.
- cmspqokmf0684jlss24yd0c6fDEPENDS_ON →
Mel-cepstral distortion quantifies perceptual difference by computing RMS deviation of mel-cepstral coefficients. Remove cepstral analysis and the coefficients cannot be derived, so the metric stops operating.
- cmrwiv1rn00a8soacg5vdpiogINSTANCE_OF →
Mel-cepstral distortion is a specific objective metric for speech quality assessment, quantifying perceptual difference via mel-cepstral coefficient RMS deviation. It is a kind of metric in the measurement-theoretic sense: a defined function mapping utterance pairs to a real-valued distance.
- cmspg0fzk056ijlss3ydafq7bDEPENDS_ON →
Mel-cepstral-distortion operates on mel-frequency spectral representations of speech signals. Remove spectral-analysis (the decomposition of signals into frequency components) and the mel-cepstral coefficients cannot be extracted. The metric fundamentally depends on spectral analysis to produce its input data.
- audio-processingSERVES →
Mel-cepstral distortion is a metric built and maintained for the sake of audio processing practice: it quantifies spectral distortion between reference and synthesized speech for voice conversion and codec evaluation, used in research and product testing. Servant→master per Law 8d.
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Sep 2, 2026, 2:47 AM UTC
- Content hash
- ec3377e068b58197ca72bdb6ff6a1d4b7180b8f83f9d26efd074a0b8c64e07e7