Spectral Envelope Representation using Sums of Gaussians

Lecture Notes in Computer Science, 2006

In this paper, a novel pitch mean based frequency warping (PMFW) method is proposed to reduce the pitch variability in speech signals at the frontend of speech recognition. The warp factors used in this process are calculated based on the average pitch of a speech segment. Two functions to describe the relations between the frequency warping factor and the pitch mean are defined and compared. We use a simple method to perform frequency warping in the Mel-filter bank frequencies based on different warping factors. To solve the problem of mismatch in bandwidth between the original and the warped spectra, the Mel-filters selection strategy is proposed. At last, the PMFW melfrequency cepstral coefficient (MFCC) is extracted based on the regular MFCC with several modifications. Experimental results show that the new PMFW MFCCs are more distinctive than the regular MFCCs.

Applied Sciences, 2016

The aim of the present work is to perform a step towards more natural pitch shifting techniques in singing voice for its application in music production and entertainment systems. In this paper, we present an advanced method to achieve natural modifications when applying a pitch shifting process to singing voice by modifying the spectral envelope of the audio excerpt. To this end, an all-pole model has been selected to model the spectral envelope, which is estimated using a constrained non-linear optimization. The analysis of the global variations of the spectral envelope was carried out by identifying changes of the parameters of the model along with the changes of the pitch. With the obtained spectral envelope transformation functions, we applied our pitch shifting scheme to some sustained vowels in order to compare results with the same transformation made by using the Flex Pitch plugin of Logic Pro X and pitch synchronous overlap and add technique (PSOLA). This comparison has been carried out by means of both an objective and a subjective evaluation. The latter was done with a survey open to volunteers on our website.

Spectral based analysis/synthesis techniques offer powerful tools for the processing of sounds. They include many different approaches, each one with its own advantages and set of compromises that have to be considered before using any of them for a particular application. In this article we present several complementary spectral representations, techniques to obtain them or generate sound from them, and a recently developed graphical interface to better explore their capabilities.

Proc. ICA …, 2010

A new spectral estimation method which improves processed sound quality of STRAIGHT, a speech analysis, modification and re-synthesis framework widely used for high-quality speech and singing manipulations, is proposed. Application of the proposed method to TANDEM-STRAIGHT, a completely reformulated version of STRAIGHT, yielded the best spectral envelope approximation among conventional methods such as LPC, cepstrum and legacy-STRAIGHT. TANDEM-STRAIGHT consists of two parts, a temporarily stable power spectrum estimation method of periodic signals (TANDEM) and a spectral envelope calculation method based on consistent sampling theory. The proposed method uses F0-adaptive smoothing and compensation of logarithmic power spectrum, for improving approximation accuracy of spectral peaks, which effects on the quality of re-synthesized sound. A series of simulations was conducted to optimize internal parameters of the proposed method. The optimized system was evaluated and compared with conventional methods using stylized spectra and simulated speech spectra. The evaluation was based on a spectral distance measure proposed by Itakura and Saitou with modification to perceptually relevant ERB N number frequency axis. The following set of spectra were used. Power spectra calculated from vocal tract area functions measured using MRI data with LF-model excitation spectra were used as the grand truth and spectral distances between this target and the estimated spectra were evaluated. A set of periodic pulse train was used for excitation signal in this case. These evaluation results indicated that the proposed method yields the smallest spectrum distance among conventional methods such as LPC, cepstrum and legacy-STRAIGHT.

2001

Abstract When designing audio effects for music processing, we are always aiming at providing higherlevel representations that may somehow fill in the gap between the signal processing world and the end-user. Spectral models in general, and the Sinusoidal plus Residual model in particular, can sometimes offer ways to implement such schemes.

… , Speech, and Signal …, 2004

This paper presents a method for modelling the speech spectral envelope using a mixture of Gaussians (MOG). A novel variational Bayesian (VB) framework for Gaussian mixture modelling of a histogram enables the derivation of an objective function that can be used to simultaneously optimise both model parameter distributions and model structure. A histogram representation of the STRAIGHT spectral envelope, which is free of glottal excitation information, is used for parametrisation using this MOG model. This results in a parameterisation scheme that purely models the vocal tract resonant characteristics. Maximum likelihood (ML) and variational Bayesian (VB) solutions of the mixture model on histogram data are found using an iterative algorithm. A comparison between ML-MOG and VB-MOG spectral modelling is carried out using spectral distortion measures and mean opinion scores (MOS). The main advantages of VB-MOG highlighted in this paper include better modelling using fewer Gaussians in the mixture resulting in better correspondence of Gaussians and formant-like peaks, and an objective measure of the number of Gaussians required to best fit the spectral envelope.

Proceedings of the SMC Conferences, 2013

Source-filter modeling of musical tones requires a filter model for the spectral envelope of the signal. Since the perceptual frequency resolution is best at low frequencies, frequency warping has been previously shown to improve spectral envelope estimation of audio signals. In this paper, considering low-order modeling for harmonic tones, we investigate the perceptual performance of three warped models which extend the filter models: Linear Prediction Coding (LPC), True-Envelope based Linear Prediction (TE-LPC), and Discrete All-Pole method (DAP). The respective warped methods allow a continuous control of the warping factor, and here we are interested in the perceptual quality of the envelope estimation according to the warping factor for all methods. Results of our listening tests show that the frequency warping which best approximates the Bark scale, does not always give the best results.

Zenodo (CERN European Organization for Nuclear Research), 2018

The spectral envelope of a speech signal encodes information about the characteristics of the speech source. As a result, spectral envelope modeling is a central task in speech applications, where tracking temporal transitions in diphones and triphones is essential for efficient speech synthesis and recognition algorithms. Temporal changes in the envelope structure are often derived from estimated formant tracks, an approach which is sensitive to estimation errors. In this paper we propose a speech source model which estimates frequency and amplitude movements in the spectral envelopes of speech signals and does not rely on formant tracking. The proposed model estimates the amplitude and frequency shifts for each sub-band and time frame of a speech signal using the information from the previous time frame. Our experiments demonstrate that the model captures temporal structures of spectral envelopes with high precision. The proposed model can thus be applied as an accurate loworder representation of temporal dynamics in speech spectral envelopes.

2016

We present an advanced method to achieve natural modifications when applying a pitch shifting process to singing voice by modifying the spectral envelope of the audio excerpt. To this end, an all-pole spectral envelope model has been selected to describe the global variations of the spectral envelope with the changes of the pitch. We performed a pitch shifting process of some sustained vowels with the envelope processing and without it, and compared both by means of a survey open to volunteers in our website.

2006

A novel linear transform (LT) is proposed for frequency warping (FW) with standard filterbank based MFCC features. Here, we use the idea of spectral interpolation of [9] to perform a continuous warping in the log filterbank output domain, and incorporate both interpolation and warping into a single warped IDCT matrix. The new transformation matrix is thus mathematically simpler than in , and no modification of standard MFCC feature extraction is required like the previous approach. In VTLN experiments with maximum likelihood score (MLS) estimation of the FW parameter, the new LT outperformed regular VTLN implemented by warping the Mel filterbank. In speaker adaptation experiments using the new LT to transform HMM means, the results were significantly better than MLLR for limited adaptation data and comparable to those in [8], while using the computationally simpler MLS FW estimation.