Voice source estimation for artificial bandwidth extension of telephone speech

Abstract

Artificial bandwidth extension (ABWE) of speech signals aims to estimate wideband speech (50 Hz -7 kHz) from narrowband signals (300 Hz -3.4 kHz). Applying the source-filter model of speech, many existing algorithms estimate vocal tract filter parameters independently of the source signal. However, many current methods for extending the narrowband voice source signal are limited to straightforward signal processing techniques which are only effective for high-band estimation. This paper presents a method for ABWE that employs novel data-driven modelling and an existing spectral mirroring technique to estimate the wideband source signal in both the high and low extension bands. A state-of-the-art Hidden Markov Model-based estimator evaluates the temporal and spectral envelopes in the missing frequency bands, with which the ABWE speech signal is synthesized. Informal listening tests comparing two existing source estimation techniques and two permutations of the proposed approach show an improvement in the perceived bandwidth of speech signals, in particular towards low frequencies. Subjective tests on the same data show a preference for the proposed techniques over the existing methods under test.

Bandwidth Extension of Speech Signals: A Comprehensive Review

“…In [25], proposed estimating the wideband voice source signal from the narrowband signal to extend the excitation of voiced speech. The technique was found to be especially effective in the low frequency range.…”

Sinusoidal-Based Lowband Synthesis for Artificial Speech Bandwidth Extension

“…Furthermore, a residual signal can be obtained by harmonic modeling [14]. Another approach leading to an estimated LB residual signal is to perform GMMbased classification among pretrained WB residual signals and subsequent lowpass filtering (known as modified data-driven voice source modeling) [39].…”

“…In [37] the LB spectral envelope is modeled by an autoregressive movingaverage (ARMA)-filter. Employing statistical models, a GMMbased envelope estimation scheme was presented in [40] and furthermore an HMM estimating the temporal envelope and adapting the estimated residual signal by gain manipulation was presented in [39], implicitly also defining the resulting LB spectral envelope. Using a statistical model for regression, a conventional neural network for direct LB spectral envelope estimation was developed [36].…”

Impact of hearing impairment on fricative intelligibility for artificially bandwidth-extended telephone speech in noise

“…For a low-band extension, however, the pitch of the given speaker needs to be reconstructed accurately to avoid annoying artifacts [9]. Unfortunately, pitch estimation -particularly in noise -is still quite challenging [7,10]. Due to the small dimension of their loudspeakers, mobile devices cannot sufficiently represent the lower frequencies anyway.…”

“…Some ABWE techniques extend both spectral gaps, i.e., the upper and lower one [7,8]. For a low-band extension, however, the pitch of the given speaker needs to be reconstructed accurately to avoid annoying artifacts [9].…”