Telecommunication service providers have to guarantee acceptable speech quality during a phone call to avoid a negative impact on the users’ quality of experience. Currently, there are different speech quality assessment methods. ITU-T Recommendation G.107 describes the E-model algorithm, which is a computational model developed for network planning purposes focused on narrowband (NB) networks. Later, ITU-T Recommendations G.107.1 and G.107.2 were developed for wideband (WB) and fullband (FB) networks. These algorithms use different impairment factors, each one related to different speech communication steps. However, the NB, WB, and FB E-model algorithms do not consider wireless techniques used in these networks, such as Multiple-Input-Multiple-Output (MIMO) systems, which are used to improve the communication system robustness in the presence of different types of wireless channel degradation. In this context, the main objective of this study is to propose a general methodology to incorporate wireless network parameters into the NB and WB E-model algorithms. To accomplish this goal, MIMO and wireless channel parameters are incorporated into the E-model algorithms, specifically into the $I_{e,eff}$ and $I_{e,eff,WB}$ impairment factors. For performance validation, subjective tests were carried out, and the proposed methodology reached a Pearson correlation coefficient (PCC) and a root mean square error (RMSE) of 0.9732 and 0.2351, respectively. It is noteworthy that our proposed methodology does not affect the rest of the E-model input parameters, and it intends to be useful for wireless network planning in speech communication services.

Non-negative matrix factorization (NMF) has recently been applied to temporal decomposition (TD) of speech spectral envelopes represented by line spectral frequencies. A couple of inherent TD constraints, which are otherwise handled as ad hoc exceptions, has also been incorporated using NMF, including line spectral frequency (LSF) ordering and monotonic event functions. Here, these constraints are analyzed and a third inherent constraint is incorporated into an NMF analysis. This is the complementarity in the sense that two overlapping event functions uniformly add up to one, which has been handled, at best, by a quadratic penalty term. We propose the use of an augmented Lagrangian including a term with the Lagrange multipliers (LMs). Additionally, a multiplicative update rule for the LMs is proposed, which fits nicely into the nature of NMF updates. Further, previous difficulties with nonsmooth spectral envelopes have been resolved by obtaining the spectral envelopes from TANDEM-STRAIGHT spectrograms. Good results are reached at the tight event rate of 12.3 ev/s, featuring mean log-spectral distortions ranging from 1.2 dB to about 1.5 dB depending on the regularizations.

Vector quantization (VQ) of speech spectral vectors has been improved by techniques such as split VQ (SVQ), vector transforms and direction switching. This letter proposes Intra-Predictive Switched SVQ (IPSSVQ) with direction switching by a Gaussian Mixture Model (GMM), using at the frame level the prediction-based lower-triangular transform (PLT), which has lower complexity than the Karhunen-Loève transform (KLT). It is shown that equivalent results to GMM KLT SSVQ may be obtained in the quantization of line spectral frequency (LSF) vectors from wideband speech signals, such as transparent coding throughout the range from 46 bit/frame to 41 bit/frame, with about three-fourths as much operational complexity.

Starting from the Durbin algorithm in polynomial space with an inner product defined by the signal autocorrelation matrix, an isometric transformation is defined that maps this vector space into another one where the Levinson algorithm is performed. Alternatively, for iterative algorithms such as discrete all-pole (DAP), an efficient implementation of a Gohberg-Semencul (GS) relation is developed for the inversion of the autocorrelation matrix which considers its centrosymmetry. In the solution of the autocorrelation equations, the Levinson algorithm is found to be less complex operationally than the procedures based on GS inversion for up to a minimum of five iterations at various linear prediction (LP) orders.

Linear prediction (LP) analysis, split in two stages, is proposed for a combined time-frequency analysis. The first-stage LP is used to obtain the residual signal and extract each one of its cycles, whose harmonic spectrum is then modeled by the second-stage estimate from discrete all-pole algorithms. Thus, harmonic cycle spectra are modeled with less than 1 dB in log spectral distortion (SD). Further, a method is proposed to approximate the log SD target. A linear approximation to the log power spectral ratio in the log SD gradient is shown to provide better model fit to harmonic cycle spectra.

Waveform interpolation (WI) models for speech coding contain many parameters whose sampling rates may not be simply related so that most implementations tend to fix their rates right from the waveform extraction stage, thereby compromising quality by departing from perfect reconstruction. A waveform extractor is proposed which samples waveform cycles of the original prediction residual signal at their natural variable rate so that it can perfectly reconstruct the signal. The speech coder, which may operate at a uniform sampling rate, is coupled to the waveform extractor by means of an evolving waveform interpolator that may handle several interpolation methods and sampling rates for a variety of fixed and variable rate coders, including conventional WI coders.

Sparse excitation from fixed codebooks provides for efficient search algorithms and high-quality speech. However, as the bit rate decreases, sparsity causes hearing effects that may be phase-compensated. A procedure is described for training the impulse response of a phase compensation filter over a large database. Two four-pulse ACELP (algebraic code-excited linear predictive) coders operating at 5.3 kbit/s have their compensation filters trained. One of the coders searches the fixed codebook with the focused algorithm and the other one uses the joint position and amplitude search (JPAS) algorithm. Listening tests were performed involving both compensated and uncompensated versions of these coders at 5.3 kbit/s and the uncompensated version at 8 kbit/s. The lower rate coders provided with trained sparsity compensation are perceived to deliver speech quality equivalent to that achieved with 8 kbit/s coders. Additionally, the beneficial hearing effect is more pronounced with the computationally efficient joint position and amplitude search.

A joint position and amplitude search algorithm is proposed for algebraic multipulse codebooks to be used in code-excited linear predictive (CELP) coders. The joint search complexity is below one quarter that of the focused search and ranks below those of the G.729A and IS-641-A coders. Listening tests indicate an equivalence in perceived quality.

A joint amplitude and position search procedure is proposed for searching algebraic multipulse codebooks. It is implemented within the reference G.723.1 codec as an example. This joint search method is shown to reduce down to one third the number of comparisons per subframe relative to the focused search over an extensive speech database. An efficient implementation of the joint search is derived which incorporates backward filtering of the residual target vector and precomputation of autocorrelation elements, bringing about a reduction in complexity of one-third in comparison to the focused search. The joint search performs about one thirtieth as many comparisons as the full position search.