1、I NTERNATIONAL TELECOM MUN ICATION UN ION Additions to the Handbook on Telephonometry TE LEC OMM UN I CATI ON STAN DARD I ZAT I ON SECTOR OF ITU 1999 INTERNATIONAL TE LECOMMUN ICATION UN ION Additions to the Handbook on Telephonometry TE LEC OMMUNICATION STANDARDIZATION SECTOR OF ITU 1999 O U 1999 A
2、ll rights reserved. No part of this publication may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying and microfilm, without permission in writing from the U. ADDITIONS TO THE HANDBOOK ON TELEPHONOMETRY CONTENTS Page 1 7 9 Addition to 3 2.3.6 of
3、the Handbook on Telephonometry . Addition to 3 3.2.5 of the Handbook on Telephonometry . Addition to 3 4.8 of the Handbook on Telephonometry . Additions to the Handbook on Telephonometry (1999) 111 Addition to 9 2.3.6 of the Handbook on Telephonometry SOME EFFECTS OF SIDETONE 1 Introduction Over a n
4、umber of years, sidetone has been studied and some important conclusions have been reached from the point of view of the subscriber in his role as both talker and listener. These conclusions relate to the effect of sidetone on a subscriber as he hears his own voice, the way his talking level changes
5、 as a result and some effects of sidetone when the subscriber is listening in conditions of moderate to high-level room noise. These effects are summarized in Figures 1 and 3. The relationship between talker and listener sidetone for a given telephone depends primarily on two factors: a) b) the geom
6、etry of its handset; and whether or not there are any non-linear gain or loss characteristics in the sidetone path. Some guidance for telephone set designers is provided in clause 4. Some information is also provided concerning the increasingly frequent occurrence of short delay talker echo, which m
7、ay be perceived as unpleasant talker sidetone. 2 Talker sidetone Figure 1 shows that there is a preferred range for sidetone when the subscriber is talking under quiet conditions, and that the difference between the sidetone being objectionable or too quiet is of the order of 20 dB. (These results w
8、ere obtained from talking-only tests and need to be confirmed by conversation tests.) The preferred range lies between 7 and 12 dB, STMR (Sidetone Masking Rating - Recommendation P.76 i). The acceptable range is wider and lies between an STMR of 1 dB and 17 dB (although it must be stated that increa
9、sing STMR to a value greater than 17 dB is likely to affect only the talking level, and that only marginally). This range corresponds to the difference between the two curves at the 50% appraisals level. It is not proposed that the 17 dB figure should in any way be considered a maximum value. Howeve
10、r, for an STMR above 20 dB, the connection sounds “dead”. Figure 2 shows the way in which the talking level changes with sidetone level. These results were obtained by means of conversation tests, for a connection close to the preferred overall loss. For telephone connections where the OLR is in the
11、 preferred range, the STMR values may be positioned in the preferred STMR range given above. However, on high-loss connections the STMR value should be close to, or even exceed, 12 dB to encourage the subscriber to speak louder. On low-loss connections, the STMR value may be sometimes permitted to b
12、ecome less than 7 dB, but only rarely should it become as low as 1 dB, e.g. telephone sets with receive volume control. Recommendation G.121 2 interprets those results for transmission planning purposes. The speech voltage will also be a function of room noise level. For modern telephones with linea
13、r transmitters, it may be desirable to have the STMR value in the range from 10 to 15 dB if it is expected that telephones will be used in noisy environments. 3 Listener sidetone High room noise in the subscribers environment disturbs the received speech in two ways: i) noise being picked up by the
14、handset microphone and transmitted to the handset receiver via the electric sidetone path; noise leaking past the earcap at the handset receiver. ii) Additions to the Handbook on Telephonometry (1999) 1 % 1 O0 75 ln - ._ 2 8 g 50 25 O 4 Objectionable level -20 -1 o O 10 20 30 dB Sidetone Masking Rat
15、ing (STMR) T1204110-92d01 (for a sealed ear) NOTE 1 - Conversational conditions will determine what part of this range is acceptable for a given connection. NOTE 2 - This part of the acceptable range (1 to 7 dB) should only be entered with caution, e.g. on low-loss connections (see Recommendation G.
16、121) or where there is a receive volume control. Figure 1 - Curves showing sidetone levels that are objectionable and too quiet, together With the preferred range, for the subscriber as a talker Studies have shown that at low frequencies the earcap leakage path dominates over the electric sidetone p
17、ath in much the same way as the human sidetone signal does in talker sidetone. The weightings applied in the STMR loudness calculation are therefore applicable and the Listener Sidetone Rating (LSTR, Recommendation P.76) has been developed, which makes use of the room noise sidetone sensitivity (see
18、 clause 9P.64 3) in the STMR rating method (see Recommendation P.79 4). Results of subjective tests of LSTR vs MOS (using in this case a mean opinion scale of 0-10) are given in Figure 3. In each case, the LSTR was derived by making use of Asm (see Recommendations P.10 5, P.64, P.79 and 3.3.17, Part
19、 C in the Handbook on Telephonometry 6) to convert the sidetone sensitivities SmesT to SRNS before calculating LSTR or applied as a weighted correction to STMR as described in A.4.3.3/G.111 7. Room noise levels were comparable at 55-59 dBA. Based upon these results, Recommendation G.121 recommends t
20、hat a value of 13 dB LSTR should be striven for. The value 13 dB is based on a 10 dB LSTR (which may be considered a minimum value), where no further improvement in mean opinion score was possible by increasing LSTR (see Figure 3), plus an allowance of 3 dB reflecting the fact that room noise in som
21、e office locations can exceed the values used in these experiments. Other tests have also suggested that a higher figure might be more appropriate. The value that is satisfactory in a given telephone connection will depend on such factors as the level of room noise, the OLR of the connection, the ta
22、lking levels used, etc. In particular, modern telephones with linear transmitters are more efficient at picking up background room noise. In this case, it may be desirable to have LSTR 15 dB. 2 Additions to the Handbook on Telephonometry (1999) dB 5 O -5 -1 o -1 5 g4 -20 -1 o O 10 20 30 40 dB T12041
23、20-92/d02 STMR (iE included) Figure 2 - Speech voltage as a function of STMR Noise levels 55-59 dBA 8 7 A 6 O 5 3 2 1 O -1 o -5 O 5 10 15 20 dB LSTR T1204130-92/dO3 X Experiment 1 O Experiment 2 A Experiment 3 Figure 3 - MOS as a function of LSTR calculated from different test results Additions to t
24、he Handbook on Telephonometry (1999) 3 4 Relationship between talker and listener sidetone 4.1 Telephones having linear sidetone characteristics For telephones having linear gain or loss characteristics in the sidetone path, the relationship between talker and listener sidetone levels is controlled
25、by the geometry of the handset. There are two aspects of the geometry that seem most important: the distance from the mouth to the transmitter port and the size of the obstacle created by the transmitter end. For speech inputs, a handset having a large transmitter end positioned close to the mouth e
26、xperiences a greater sound pressure at its transmitter port than a handset having the transmitter end positioned farther from the mouth (distance effect) or one having a small transmitter end (obstacle effect). However, for diffuse field room noise inputs, the sound pressure at the transmitter port
27、is independent of the size and shape of the handset. Thus, if the STMR level is the same for the two handsets, the one with the large transmitter end close to the mouth will have less electrical gain in its sidetone path, which will result in a greater LSTR value. It has been shown that the differen
28、ce in LSTR and STMR levels for a sample of 26 linear telephone sets is highly correlated to the logarithm of the distance between the transmitter port (centre of the external opening for the microphone on the surface of the handset) and the centre of the lip ring of the artificial mouth when the han
29、dset is placed in the LRGP test position (see Recommendation P.64). It has the following empirical relationship: LSTR - STMR = 33 - 20log(d) where the distance d from the transmitter port to the centre of the lip ring is measured in millimetres. There may be small perturbations in the order of k 1 d
30、B about this relationship depending on the obstacle size presented by the transmitter end of the handset. NOTE - This relationship is based on measurements of telephones having more or less conventional handsets. It may not be applicable for handsets with extreme geometries or for operator headsets
31、that have their transmitter ports located behind the lip plane. 4.2 Telephones having non-linear sidetone characteristics Non-linear gain or loss characteristics may be used in the electrical sidetone path to increase the LSTR - STMR difference. Carbon transmitters, for example, frequently are less
32、sensitive to the lower input levels of room noise than they are to the higher input levels of speech. Such a characteristic may be introduced into telephones having linear microphones through the use of various non-linear gain circuits. If the same non-linear gain function is used in both the send a
33、nd sidetone paths of the telephone, then the LSTR - STMR difference may be approximated by measuring the difference in send sensitivities due to speech and room noise inputs, DELSM, as described in Recommendation P.64. A STMR difference may then be calculated according to the method given in Annex A
34、/G. 11 1. However, if the send and sidetone paths do not have the same non-linear gain characteristics (e.g. automatic gain control circuitry in the receive path that affects sidetone), then the DELSM method will give erroneous results. In this case, the LSTR and STMR values must be measured directl
35、y. 5 Short delay talker echo perceived as sidetone Talker echo can have a detrimental effect on transmission quality at delay times of a few milliseconds, even though the delay is not long enough for it to be perceived as an echo signal separate from the sidetone. Such echoes can occur, for example,
36、 due to reflections from the analogue trunk port of a digital PBX or on local analogue calls through a digital exchange. Unless the hybrid that converts the 4-wire digital PBX or exchange back to a 2-wire analogue circuit is well matched, some reflections will occur. Because of the digital processin
37、g times involved, these talker echo signals have a few milliseconds of delay. Sidetone provides a beneficial masking of low-level short delay talker echo, but as the talker echo level increases, it interacts with the sidetone in an unpleasant manner (hollow-sounding sidetone, rain-barrel effect, etc
38、.). 4 Additions to the Handbook on Telephonometry (1999) The objectively measurable effect of short delay talker echo is that it produces ripples in the sidetone frequency response. The reflected talker echo signal is added to the direct sidetone signal with a phase relationship that increases the s
39、ignal at some frequencies and decreases it at others. The spacing between the ripples is equal to the reciprocal of the delay. When the reflected talker echo signal is small relative to the direct sidetone, the ripples are small. As the talker echo signal increases in magnitude, the ripples increase
40、 in size until the peaks are 6 dB above the in-phase signal and the troughs are very deep due to almost exact out-of-phase cancellation. At even higher talker echo levels (or lower sidetone levels) the amount of ripple again decreases, but the predominant signal is then the delayed talker echo. Peop
41、le perceive short delay talker echo combined with sidetone differently than an equivalent level of pure sidetone, even though they may not be able to detect that a separate echo signal is present. Thus, a simple sidetone measure such as STMR is not adequate to describe the effect of the combined sig
42、nal. Talker echo, even with very short delay times, must be treated as a separate impairment to transmission quality. Recommendation P.11 8 and reference 9 provide some guidance as to how both sidetone and talker echo may be taken into account in predicting the quality of a telephone connection, but
43、 this subject remains under study. References il CCT Recommendation P.76 (1988), Determination of loudness ratings; findamental principles. Pl U-T Recommendation G. 121 (1993), Loudness Ratings (LRs) of national systems. 31 U-T Recommendation P.64 (1997), Determination of sensitivity/equency charact
44、eristics of local telephone systems. 41 U-T Recommendation P.79 (1993), Calculation of loudness ratings for telephone sets. 51 U-T Recommendation P. 10 (1998), Vocabulary of terms on telephone transmission quality and telephone sets. 61 Handbook on telephonometry, U, Geneva, 1993. 71 U-T Recommendat
45、ion G. 11 1 (1993), Loudness ratings (LRs) in an international connection. SI U-T Recommendation P. 11 (1993), Effect of transmission impairments, Pl Addition to 3 3.2.5 of the Handbook on Telephonometry: Artificial conversational speech, U, Geneva, 1999. Additions to the Handbook on Telephonometry
46、(1999) 5 Addition to 9 3.2.5 of the Handbook on Telephonometry ARTIFICIAL CONVERSATIONAL SPEECH The Artificial Voice described in Recommendation P.50 is used as a one-way test signal in the performance evaluation of, for example, low bit-rate speech codecs. It is required that the application area o
47、f these artificial signals be enlarged for the performance evaluation of devices that are operated by speech signals, such as echo cancellers and voice switches in loudspeaker telephone sets and DSI (Digital Speech Interpolation) devices. For these purposes, artificial signals should simulate conver
48、sations by humans. Therefore, they should contain not only “talkspurt” periods expressed in terms of the Artificial Voices in Recommendation P.50 but also “pause” periods. Artificial signals are also required to be two-way signals for simulating a “talker and listener” environment. A signal that alt
49、ernately places the Artificial Voice in Recommendation P.50 of arbitrary duration and silence (zero sequences) of arbitrary duration along a time axis is not adequate, because human conversational speech has specific characteristics as conversation. For example, humans cannot continue to utter for a long period without pause and there are few cases of two persons uttering simultaneously for a long period. Therefore, some statistical temporal characteristics of real conversational speech should be simulated in artificial signals. An artificial
