TUTORIAL ARTICLE The phase vocoder: theory and practice

21St European Signal Processing Conference, 2013

When dealing with musical sounds, the short-time Fourier transform prevails and sinusoids play a key role, according to both acoustics (vibrating modes) and psychoacoustics (pure tones). The values obtained when decomposing the signal on the time-frequency atoms are usually assigned to their geometrical center, leading to estimation errors for the sinusoidal parameters. To correct this, one can exploit the amplitude or phase information, use the derivatives of the analysis window, or those of the audio signal. This leads to three methods (phase vocoder, spectral reassignment, derivative algorithm) equally efficient: they are in fact different formulations of the best analysis method based on the Fourier spectrum.

Computer Music Journal, 2000

IEEE Transactions on Audio, Speech, and Language Processing, 1999

The phase vocoder is a well established tool for time scaling and pitch shifting speech and audio signals via modification of their short-time Fourier transforms (STFTs). In contrast to time-domain time-scaling and pitch-shifting techniques, the phase vocoder is generally considered to yield high quality results, especially for large modification factors and/or polyphonic signals. However, the phase vocoder is also known

1987

Abstract This paper describes a peak-tracking spectrum analyzer, called Parshl, which is useful for extracting additive synthesis parameters from inharmonic sounds such as the piano. Parshl is based on the Short-Time Fourier Transform (STFT), adding features for tracking the amplitude, frequency, and phase trajectories of spectral lines from one FFT to the next.

In this paper we propose a new method to reduce phase vocoder artifacts during attack transients. In contrast to all transient preservation algorithms that have been proposed up to now the new approach does not impose any constraints on the time dilation parameter for processing transient segments. By means of an investigation into the spectral properties of attack transients of simple sinusoids we provide new insights into the causes of phase vocoder artifacts and propose a new method for transient preservation as well as a new criterion and a new algorithm for transient detection. Both, the transient detection and the transient processing algorithms are designed to operate on the level of spectral bins which reduces possible artifacts in stationary signal components that are close to the spectral peaks classified as transient. The transient detection criterion has a close relation to the transient position and allows us to find an optimal position for reinitializing the phase spectrum. The evaluation of the transient detector by means of a hand labeled data base demonstrates its superior performance compared to a previously published algorithm. Attack transients in sound signals transformed with the new algorithm achieves high quality even if strong dilation is applied to polyphonic signals.

Fourth Meeting of the Fwo Research Society on Foundations of Music Research, 1998

The phase vocoder has been used as a time-scale-modification tool for several decades. Applying large positive modification factors to different kinds of sounds (time-stretching), the result will always sound "phasy" or "reverberant". The sound quality can be improved by "locking" the phases. Phase-locking preserves the phase relations around a local maximum in the magnitude spectrum. For large modification factors, locking the entire phase spectrum sounds "rigid". In this work, the deterministic and stochastic components of a sound are separated in the frequency domain, and only the phases of sinusoids are locked, while the remaining phases are set to random numbers. The deterministic part is detected within a "reduced variance" magnitude spectrum using heuristic conditions. The reduced variance spectrum is calculated by weighting the actual magnitude spectrum, the spectrum of a frame about 11 ms in advance and the reduced variance spectrum 11 ms before the actual spectrum. In this way, the variance of the approximation can be reduced for the peak-detection yielding better results for noise-like signals. For resynthesis, the deterministic magnitudes are combined with the smoothed stochastic magnitudes, and the locked phases are combined with the random phases.

1999

The phase-vocoder is usually presented as a high-quality solution for time-scale modification of signals, pitch-scale modifications usually being implemented as a combination of timescaling and sampling rate conversion. We present two new phase-vocoder-based techniques which allow direct manipulation of the signal in the frequency-domain, enabling such applications as pitch-shifting, chorusing, harmonizing, partial stretching and other exotic modifications which cannot be

IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017

The phase-vocoder is a well-known tool for the frequency domain processing of speech or audio signals, with applications such as time compression or expansion, pitch-scale modification, noise reduction, etc. In the context of time-scale or pitch-scale modification, the phase-vocoder is usually considered to yield high quality results, especially when large modification factors are used on polyphonic or non-pitched signals. However, the phase-vocoder is also known for an artifact that plagues its output, and has been described in the literature as either “phasiness”, “reverberation”, or “loss of presence”. Research has been devoted to understanding and reducing this artifact, and solutions have been proposed which either significantly improve the quality of the output at the cost of a very high additional computation time, or are inexpensive but only marginally effective. This paper examines the problem of phasiness in the context of time-scale modification of signals, and presents t...