Cepstrum and Liftering Parameter Calculator

Find a signal's pitch quefrency and recommended liftering cutoff for cepstral analysis.

🎙️ Cepstrum and Liftering Parameter Calculator
Sampling rate (fs)8000
Hz
4,000 Hz192,000 Hz
Fundamental frequency (f0)100
Hz
50 Hz1,000 Hz
Liftering cutoff fraction (α)0.5
×N₀
0.10.9
Pitch quefrency
Pitch period (time)
Liftering cutoff (samples)
Liftering cutoff (time)
Step-by-step working

🎙️ What is Cepstrum and Liftering?

The cepstrum and liftering together form a classic signal processing technique for separating a signal's slowly varying spectral envelope from its fast, periodic excitation structure, most famously used to isolate a speaker's vocal tract shape from their pitch. Taking the inverse Fourier transform of the log-magnitude spectrum turns what was a multiplication of spectra (or a convolution in time) into a simple sum in a new domain called quefrency, so the two components can be separated with an ordinary low-pass or high-pass filter, called a lifter, applied in that domain.

Speech scientists and voice coding engineers use cepstral liftering to separate vocal tract formants (low quefrency) from glottal pitch pulses (high quefrency), a core step in homomorphic vocoders, MFCC feature extraction for speech recognition, and pitch (fundamental frequency) tracking. Audio and music information retrieval systems use the same cepstral pitch peak detection to estimate a note's fundamental frequency independent of its timbre. Seismologists and radar engineers apply the identical mathematics (originally the field cepstral analysis was invented for) to detect echoes and layered reflections buried in a signal's spectrum.

A common misconception is that liftering is just another name for ordinary frequency-domain filtering. It operates in an entirely different domain, quefrency, not frequency, precisely because the whole point of the cepstrum is to convert a convolution (like vocal tract filtering applied to a pitch source) into an additive combination that a simple threshold-based filter can separate, something impossible directly in the frequency domain.

This calculator computes a signal's pitch quefrency directly from its sampling rate and fundamental frequency, then recommends a liftering cutoff quefrency based on an adjustable fraction of that pitch period.

📐 Formula

N0  =  fs / f0     nc  =  α×N0
fs = sampling rate (Hz)
f0 = fundamental (pitch) frequency (Hz)
N0 = pitch quefrency (samples)
α = liftering cutoff fraction (0 to 1)
Example: fs = 8,000 Hz, f0 = 100 Hz, α = 0.5 → N0 = 80 samples (10 ms), nc = 40 samples (5 ms).

📖 How to Use This Calculator

Steps

1
Enter the sampling rate and fundamental frequency. Type in the signal's sampling rate and its fundamental (pitch) frequency, both in Hz.
2
Choose a liftering cutoff fraction. Set the fraction of the pitch period used to place the liftering cutoff (0.5 is a common default).
3
Read the quefrency and liftering cutoff. Click Calculate to see the pitch quefrency and the recommended liftering cutoff quefrency, in samples and time.

💡 Example Calculations

Example 1 — Low-Pitched Male Voice at 8 kHz

fs = 8,000 Hz, f0 = 100 Hz, α = 0.5

1
N0 = 8,000/100 = 80.00 samples (10.0000 ms)
2
nc = 0.5×80.00 = 40.00 samples (5.0000 ms)
Pitch quefrency = 80.00 samples, liftering cutoff = 40.00 samples
Try this example →

Example 2 — Higher-Pitched Voice at 16 kHz

fs = 16,000 Hz, f0 = 200 Hz, α = 0.5

1
N0 = 16,000/200 = 80.00 samples (5.0000 ms)
2
nc = 0.5×80.00 = 40.00 samples (2.5000 ms)
Pitch quefrency = 80.00 samples, liftering cutoff = 40.00 samples
Try this example →

Example 3 — CD-Quality Musical Note (A4, 441 Hz)

fs = 44,100 Hz, f0 = 441 Hz, α = 0.6

1
N0 = 44,100/441 = 100.00 samples (2.2676 ms)
2
nc = 0.6×100.00 = 60.00 samples (1.3605 ms)
Pitch quefrency = 100.00 samples, liftering cutoff = 60.00 samples
Try this example →

❓ Frequently Asked Questions

What is the cepstrum?+
The cepstrum is the inverse Fourier transform of the logarithm of a signal's power spectrum, and it converts a signal that is a convolution in the time domain (like a vocal tract filter applied to a pitch excitation) into a sum in the cepstral domain, making the two components separable by simple filtering (liftering).
What is quefrency?+
Quefrency is the cepstral domain's equivalent of time, an anagram-styled term coined alongside 'cepstrum' itself. A signal's pitch period appears as a sharp peak in the cepstrum at the quefrency equal to that period (in samples or seconds), which is exactly how cepstral pitch detection works.
How do I find the pitch quefrency from sampling rate and fundamental frequency?+
Pitch quefrency in samples is N0 = fs / f0, where fs is the sampling rate and f0 is the fundamental (pitch) frequency. An 8 kHz-sampled voice with a 100 Hz fundamental has a pitch quefrency of 80 samples (10 ms), matching the signal's actual pitch period.
What is liftering?+
Liftering is filtering performed in the cepstral (quefrency) domain, the cepstral analog of ordinary time-domain or frequency-domain filtering. Low-time liftering keeps only low-quefrency cepstral coefficients to recover a smooth spectral envelope (the vocal tract shape), while high-time liftering isolates the higher-quefrency excitation (pitch) structure.
How do I choose a liftering cutoff quefrency?+
The liftering cutoff should sit below the pitch quefrency N0 but above the low quefrencies that represent the vocal tract envelope, a common rule of thumb sets the cutoff to roughly 40 to 60 percent of the expected pitch period, low enough to exclude pitch harmonics but high enough to retain formant envelope detail.
Why does the liftering cutoff need to change with fundamental frequency?+
Because pitch quefrency N0 = fs/f0 shifts inversely with f0, a fixed absolute cutoff (in samples) that works for a low-pitched voice can accidentally include (or exclude) pitch information for a high-pitched voice. Expressing the cutoff as a fraction of the (estimated or measured) pitch period keeps the envelope/excitation separation consistent across different speakers or pitches.
How is cepstral pitch detection used in practice?+
Cepstral pitch detection searches the cepstrum's high-quefrency region (above the liftering cutoff) for its highest peak, and the quefrency location of that peak directly gives the pitch period. This method is robust to spectral envelope shape and formant structure, which is why it remains a standard technique in speech pitch (F0) trackers alongside autocorrelation-based methods.
What is homomorphic vocoding?+
Homomorphic vocoding is a speech coding and synthesis technique that separates a voice signal into its vocal tract envelope (via low-time liftering of the cepstrum) and its pitch excitation (via high-time liftering), transmitting or storing the two components separately, since together they can be recombined to reconstruct the original speech, similar in spirit to how a classic LPC vocoder separates formants from pitch.
Does the fundamental frequency need to be below the Nyquist frequency?+
Yes. The fundamental frequency f0 must be below fs/2 (the Nyquist frequency) for the sampled signal to represent that periodicity at all, otherwise the pitch itself would alias and no valid quefrency peak would appear at the expected location.
Is quefrency measured in samples or seconds?+
Both are used interchangeably, quefrency in samples (n) converts to quefrency in time (seconds) by dividing by the sampling rate, t = n/fs, exactly analogous to converting a sample index into a time value in the ordinary time domain. This calculator reports both forms for the pitch quefrency and the liftering cutoff.

What is the cepstrum?

The cepstrum is the inverse Fourier transform of the logarithm of a signal's power spectrum, and it converts a signal that is a convolution in the time domain (like a vocal tract filter applied to a pitch excitation) into a sum in the cepstral domain, making the two components separable by simple filtering (liftering).

What is quefrency?

Quefrency is the cepstral domain's equivalent of time, an anagram-styled term coined alongside 'cepstrum' itself. A signal's pitch period appears as a sharp peak in the cepstrum at the quefrency equal to that period (in samples or seconds), which is exactly how cepstral pitch detection works.

How do I find the pitch quefrency from sampling rate and fundamental frequency?

Pitch quefrency in samples is N0 = fs / f0, where fs is the sampling rate and f0 is the fundamental (pitch) frequency. An 8 kHz-sampled voice with a 100 Hz fundamental has a pitch quefrency of 80 samples (10 ms), matching the signal's actual pitch period.

What is liftering?

Liftering is filtering performed in the cepstral (quefrency) domain, the cepstral analog of ordinary time-domain or frequency-domain filtering. Low-time liftering keeps only low-quefrency cepstral coefficients to recover a smooth spectral envelope (the vocal tract shape), while high-time liftering isolates the higher-quefrency excitation (pitch) structure.

How do I choose a liftering cutoff quefrency?

The liftering cutoff should sit below the pitch quefrency N0 but above the low quefrencies that represent the vocal tract envelope, a common rule of thumb sets the cutoff to roughly 40 to 60 percent of the expected pitch period, low enough to exclude pitch harmonics but high enough to retain formant envelope detail.

Why does the liftering cutoff need to change with fundamental frequency?

Because pitch quefrency N0 = fs/f0 shifts inversely with f0, a fixed absolute cutoff (in samples) that works for a low-pitched voice can accidentally include (or exclude) pitch information for a high-pitched voice. Expressing the cutoff as a fraction of the (estimated or measured) pitch period keeps the envelope/excitation separation consistent across different speakers or pitches.

How is cepstral pitch detection used in practice?

Cepstral pitch detection searches the cepstrum's high-quefrency region (above the liftering cutoff) for its highest peak, and the quefrency location of that peak directly gives the pitch period. This method is robust to spectral envelope shape and formant structure, which is why it remains a standard technique in speech pitch (F0) trackers alongside autocorrelation-based methods.

What is homomorphic vocoding?

Homomorphic vocoding is a speech coding and synthesis technique that separates a voice signal into its vocal tract envelope (via low-time liftering of the cepstrum) and its pitch excitation (via high-time liftering), transmitting or storing the two components separately, since together they can be recombined to reconstruct the original speech, similar in spirit to how a classic LPC vocoder separates formants from pitch.

Does the fundamental frequency need to be below the Nyquist frequency?

Yes. The fundamental frequency f0 must be below fs/2 (the Nyquist frequency) for the sampled signal to represent that periodicity at all, otherwise the pitch itself would alias and no valid quefrency peak would appear at the expected location.

Is quefrency measured in samples or seconds?

Both are used interchangeably, quefrency in samples (n) converts to quefrency in time (seconds) by dividing by the sampling rate, t = n/fs, exactly analogous to converting a sample index into a time value in the ordinary time domain. This calculator reports both forms for the pitch quefrency and the liftering cutoff.