<title>Objective end-to-end (mouth-to-ear) conversational speech quality tests for VoIP scenarios</title>

In this paper, method for evaluation of varying conversational speech quality in wireless communications is proposed. The proposed algorithm evaluates quality degradations using indicators based on count of lost frames and voice activity indications. The correctness of proposed algorithm is investigated by comparison of test results with results obtained using PESQ algorithm under same conditions. The achieved average correlation coefficient is 0.975. This result is independent of frame loss model and percentage of silence in test sentences. Proposed algorithm can be implemented in mobile stations and used for speech quality evaluation by real conversation. Ill. 3, bibl. 13, tabl. 3 (in English; abstracts in English and Lithuanian).http://dx.doi.org/10.5755/j01.eee.109.3.182

Download Full-text

Assessment of Conversational Speech Quality Inside Vehicles, Concerning Influences of Room Acoustics and Driving Noises

Acta Acustica united with Acustica ◽

10.3813/aaa.918530 ◽

2012 ◽

Vol 98 (3) ◽

pp. 461-474 ◽

Cited By ~ 1

Author(s):

Oliver Jung

Keyword(s):

Transfer Functions ◽

Strong Relationship ◽

Speech Quality ◽

Room Acoustics ◽

Conversational Speech ◽

Acoustical Parameters ◽

Vehicle Interior ◽

Lombard Effect ◽

Speech Spectrum ◽

The Impact

This study considers the influences of room acoustics and driving noises in vehicle interiors on the subjectively perceived acoustical quality of conversations between passengers. A listening test with 25 participants was performed inside a laboratory to assess the impact of different vehicle interior transfer functions on the speech quality assessment in four predetermined dimensions. Idealized driving noises at three different vehicle speeds were presented simultaneously with speech samples to quantify the interferences of these noise conditions with varied signal-to-noise ratios. To minimize the influence of different human speakers, four talkers (two male and two female) were selected from commercially available audio books. The respective speech samples were adjusted in level and long-term average speech spectrum to the common values of conversational speech. The automatic reflex of raising one's voice in noisy environments, called “Lombard Effect” [1], was taken into account for an additional adjustment of speech levels while driving noises were present. A strong relationship between the speech-to-noise ratio and the test participants' evaluations was found. Thus, one can assume that the speech signals' attenuation or amplification caused by the different room acoustics of the tested vehicles play a more important role for a sufficient speech quality than the varied speech timbre or other parameters. Only at very high speech-to-noise ratios ( ≥ 20 dB with A-weighting), room-acoustical parameters such as IACC or the reverberation time are more determining for the speech quality appreciation than the speech's sound pressure level.

Download Full-text