Instrumental Assessment of Near-end Perceived Listening Effort

Communication in noisy situations may be extremely stressful for the person located at the near-end side. Since the background noise originates from a natural environment, it cannot be reduced for the listener. Thus, the only possibility to improve this scenario with support of digital signal processing is the insertion of speech enhancement algorithms in the downlink direction of terminals. So far no measurement technique is available to evaluate the impact of signal processing techniques such as “near-end listening enhancements” [1] (NELE), artificial bandwidth extension (BWE) or additional noise reduction (NR). For mobile phones, acoustic testing in downlink direction is always carried out in silent condition. However, in several state-of-the-art devices the aforementioned algorithms are already included. This implies that a device may behave differently under noisy conditions than in silence: e.g. NELE algorithms may be triggered by a certain noise level and/or spectrum. Whenever speech processing is inserted into a conversation, quality aspects must be considered, too. A satisfactory balance between speech quality and listening effort is desirable from the user’s point of view. Currently, no reliable objective or instrumental methods are available to evaluate speech quality and listening effort of a device under test (DUT) in downlink in the presence of background noise. Any possible metrics should take into account ongoing trends in acoustic telecommunication measurement standards, i.e.: • Usage of real speech instead of artifical test signals. • Realistic playback of background noise scenarios (e.g. according to [2] or [3]). • “Black-Box-Approach”: no internals of a DUT are known, only outer measurements are available. Due to these requirements, several existing assessment measures targeting to intelligibility and/or speech quality aspects prove to be unfavorable: • STITEL, STIPA, RASTI according to [4]: shaped noise signals are used for measurement. • ITU-T Recommendations P.862 [5], [6] and P.863 [7]: noise or near-end noise is explicitly excluded in scope. • ETSI EG 202 396-3 [8] and TS 103 106 [9]: methods are specified for noise reduction scenarios and only for uplink direction. Another widely used measure for the instrumental intelligibility assessment is the speech intelligibility index (SII) [10]. Several drawbacks of this measurement algorithm should be considered, too: • Pure 1/3 octave level-based measure, no real psychoacoustical model (except frequency weighting) • Noise-free degraded speech signal is needed as input (not available in acoustic testing) Figure 1: Recording setup for (binaural) signal assessment