Formal Subjective Tests Research Articles

Improved Signal-to-Noise Ratio Estimation for Speech Enhancement

This paper addresses the problem of single-microphone speech enhancement in noisy environments. State-of-the-art short-time noise reduction techniques are most often expressed as a spectral gain depending on the signal-to-noise ratio (SNR). The well-known decision-directed (DD) approach drastically limits the level of musical noise, but the estimated a priori SNR is biased since it depends on the speech spectrum estimation in the previous frame. Therefore, the gain function matches the previous frame rather than the current one which degrades the noise reduction performance. The consequence of this bias is an annoying reverberation effect. We propose a method called two-step noise reduction (TSNR) technique which solves this problem while maintaining the benefits of the decision-directed approach. The estimation of the a priori SNR is refined by a second step to remove the bias of the DD approach, thus removing the reverberation effect. However, classic short-time noise reduction techniques, including TSNR, introduce harmonic distortion in enhanced speech because of the unreliability of estimators for small signal-to-noise ratios. This is mainly due to the difficult task of noise power spectrum density (PSD) estimation in single-microphone schemes. To overcome this problem, we propose a method called harmonic regeneration noise reduction (HRNR). A nonlinearity is used to regenerate the degraded harmonics of the distorted signal in an efficient way. The resulting artificial signal is produced in order to refine the a priori SNR used to compute a spectral gain able to preserve the speech harmonics. These methods are analyzed and objective and formal subjective test results between HRNR and TSNR techniques are provided. A significant improvement is brought by HRNR compared to TSNR thanks to the preservation of harmonics

Subjective evaluation of MPEG-4 video codec proposals: Methodological approach and test procedures

Abstract A new audio-visual coding standard, MPEG-4, is currently under development. MPEG-4 will address not only compression, but also completely new audio-video coding functionalities related to content-based interactivity and universal access. As part of the MPEG-4 standardization process, in November, 1995 assessments were performed on technologies proposed for incorporation in the standard. These assessments included formal subjective tests, as well as expert panel evaluations. This paper describes the MPEG-4 video formal subjective tests. Since MPEG-4 addresses new coding functionalities, and also operates at bit-rates lower than ever subjectively tested before on a large scale, standard ITU test methods were not directly applicable. These methods had to be adapted, and even new test methods devised, for the MPEG-4 video subjective tests. We describe here the test methods used in the MPEG-4 video subjective tests, how the tests were carried out, and how the test results were interpreted. We also evaluate the successes and shortcomings of the MPEG-4 video subjective tests, and suggest possible improvements for future tests. The MPEG-4 video subjective tests were successful, providing the MPEG community with critical information to guide in the selection of technologies for inclusion in the video part of the MPEG-4 standard.

A mixed excitation LPC vocoder model for low bit rate speech coding

Traditional pitch-excited linear predictive coding (LPC) vocoders use a fully parametric model to efficiently encode the important information in human speech. These vocoders can produce intelligible speech at low data rates (800-2400 b/s), but they often sound synthetic and generate annoying artifacts such as buzzes, thumps, and tonal noises. These problems increase dramatically if acoustic background noise is present at the speech input. This paper presents a new mixed excitation LPC vocoder model that preserves the low bit rate of a fully parametric model but adds more free parameters to the excitation signal so that the synthesizer can mimic more characteristics of natural human speech. The new model also eliminates the traditional requirement for a binary voicing decision so that the vocoder performs well even in the presence of acoustic background noise. A 2400-b/s LPC vocoder based on this model has been developed and implemented in simulations and in a real-time system. Formal subjective testing of this coder confirms that it produces natural sounding speech even in a difficult noise environment. In fact, diagnostic acceptability measure (DAM) test scores show that the performance of the 2400-b/s mixed excitation LPC vocoder is close to that of the government standard 4800-b/s CELP coder.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Design and performance of an analysis-by-synthesis class of predictive speech coders

The performance of a broad class of analysis-by-synthesis linear predictive speech coders is quantified experimentally. The class of coders includes a number of well-known techniques as well as a very large number of speech coders which have not been named or studied. A general formulation for deriving the parametric representation used in all of the coders in the class is presented. A new coder, named the self-excited vocoder, is discussed because of its good performance with low complexity, and because of the insight this coder gives to analysis-by-synthesis coders in general. The results of a study comparing the performances of different members of this class are presented. The study takes the form of a series of formal subjective and objective speech quality tests performed on selected coders. The results of this study lead to some interesting and important observations concerning the controlling parameters for analysis-by-synthesis speech coders.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Transform coding of audio signals using perceptual noise criteria

A 4-b/sample transform coder is designed using a psychoacoustically derived noise-making threshold that is based on the short-term spectrum of the signal. The coder has been tested in a formal subjective test involving a wide selection of monophonic audio inputs. The signals used in the test were of 15-kHz bandwidth, sampled at 32 kHz. The bit rate of the resulting coder was 128 kb/s. The subjective test shows that the coded signal could not be distinguished from the original at that bit rate. Subsequent informal work suggests that a bit rate of 96 kb/s may maintain transparency for the set of inputs used in the test.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Subjective Effects of Variable Delay and Speech Clipping in Dynamically Managed Voice Systems

The purpose of this paper is to examine speech impairments likely to arise in dynamically managed voice (DMV) systems. DMV systems utilize speech activity detection to exploit speech idle time and variable bit rate coding to exploit nonstationary speech statistics. The emphasis here is on systems using speech detection. This processing introduces two impairments not commonly found in traditional communication systems variable Speech burst delay and speech clipping. Simulations of these impairments were implemented, and formal subjective testing was performed to assess subjects' reactions to a range of impairment levels. Emphasis was on formal subjective listening tests and customer opinion of speech quality as defined by a rating scale. The test conditions are applicable to general telephony, where relatively high speech quality is required. Results on variable speech burst delay and front-end and midspeech burst clipping are presented. These results serve as input to the design process and to the establishment of performance guidelines for DMV systems.

A note on complexity reduction for linear predictive speech synthesis

Most linear predictive speech synthesis (LPSS) algorithms employ a deemphasis filter with a fixed transfer function subsequent to the time varying LPSS filter. The excitation signal for the LPSS filter is usually a stored waveform sequence, repeated at the pitch period, for voiced sounds or a noise sequence for Unvoiced sounds. In either case the excitation signal is approximately white. A computational savings may be achieved by using an excitation signal whose spectrum reflects the frequency response of the deemphasis filter. A procedure is described for generating such an excitation signal and formal subjective intelligibility test results are presented to demonstrate that no performance degradation occurs when this method is employed.