1、 Access to Additional Content for TIA/EIA-125-A, Dated: August 2000 (Click here to view the publication) This Page is not part of the original publication This page has been added by IHS as a convenience to the user in order to provide access to additional content as authorized by the Copyright hold
2、er of this document Click the link(s) below to access the content and use normal procedures for downloading or opening the files. TIA-EIA-125-A Files Information contained in the above is the property of the Copyright holder and all Notice of Disclaimer a fair implementation of the algorithm in soft
3、ware; or some other mechanism. A fair3implementation in software shall yield bit exact output with reference to any hardware implementation that it is4claimed to represent.5The input speech material has been precision-limited by an 8-bit -law quantization algorithm in which the6inverse quantized lin
4、ear samples fill the entire 16-bit linear range. As specified in 2.4.2.1.1 of IS-96-C, the master7codec assumes a 14-bit integer input quantization. Thus the master encoder scales down the 16-bit linear8samples by 4.0 before processing, and the master decoder scales the output up by 4.0 after proces
5、sing. In order9for the test codec to match the master codec, a similar pre-processing and post-processing scaling function shall10be performed. This may be done either internally by the test codec or externally by using a scaling function on11the input and output speech. (Note: one method of externa
6、lly scaling the data is to use the program scale.c12supplied in directory /tools of the companion software, (see 3.3.6) with a 0.25 factor before encoder13processing and a 4.0 factor after decoder processing.)14TIA/EIA/125-A1-6No text.1TIA/EIA/I125-A2-12 CODEC MINIMUM STANDARDS1This section describe
7、s the procedures used to verify that the speech codec implementations meet minimum2performance requirements. Minimum performance requirements may be met by either:3 Meeting strict objective performance requirements (see 2.1), or4 Meeting a combination of a less strict objective performance requireme
8、nts and subjective performance5requirements (see 2.2).6In the first case, objective performance requirements have to be met by both the encoder and decoder. In the7latter case the decoder has to meet objective performance requirements for voice quality, while the encoder has8to meet objective measur
9、es for encoding rate. Manufacturers are cautioned that the equivalent of a full floating-9point implementation of the IS-96-C codec may be required to meet the performance requirements of the strict10objective test. Minimum performance may be demonstrated, by meeting the requirements of either 2.1,
10、or those11of 2.2.122.1 Strict Objective Performance Requirements13An objective measure of accuracy is obtained by computing the segmental signal-to-noise ratio (SNRSEG)14between various encoder/decoder combinations. The specific measure of accuracy is based upon the15distribution of SNR measurements
11、 collected over a large database.16The strict objective performance test may include up to four stages (see Figure 2.1-1).17In the first stage, the speech source material as described in 3.5 is input to both the master and test encoders.18All speech source material supplied in the directory /expt1/o
12、bjective, as described in Figure 3.6-1 of the19companion software distribution, is to be used. The resulting output bit streams are saved. The output bit20stream of the master encoder is input to both the test and master decoders. The output speech signals are21compared using the procedure described
13、 in 2.1.1. If these signals agree according to the specific minimum22performance requirements for the method defined in 2.1.1, they are considered to be in close agreement. If these23signals are not in close agreement, then the strict objective performance requirements of this section are not met24(
14、see 2.2.1.3).25If there is close agreement, then the second stage of testing may proceed. The impaired bit stream from the26master encoder, found in the file /expt2/objective/expt2obj.pkt, described in Figure 3.6-1, is input27to both the test and master decoders. The output speech signals are compar
15、ed using the procedure described in282.1.1. If these signals are not in close agreement, then the strict objective performance requirements of this29section are not met (see 2.2.1.3).30TIA/EIA/IS-125-A2-21Input speech into both test and master encoders. Input impaired channel packets from master cod
16、ec into both test and master decoder.NYNStrict objective performance reqts are not met.Does master/test pass decoder requirements?Stages 1 and 2Is bitstream of test encoder equal to bitstream of master encoder?PassYStage 3Do test/master and test/test meet encoder/decoder requirements?Stage 4NNPassY2
17、Figure 2.1-1. Flow Chart for Strict Objective Performance Test3TIA/EIA/I125-A2-3If there is close agreement, then the third stage of testing may proceed. The output bit streams of the master1encoder and test encoder are compared. If the bit streams are identical, then the test encoder/decoder has me
18、t2the minimum performance requirements and no further testing is required.3If the bit streams are not identical, then the fourth stage of the strict test may be attempted. In this stage, the4speech output from the test encoder/master decoder and test encoder/test decoder combinations are both5compar
19、ed with the output of the master encoder/master decoder using the procedure described in 2.1.1. If both6combinations pass the acceptance criteria of 2.1.1, then the test encoder/decoder has met the minimum7performance requirements and no further testing is required. If both combinations do not pass
20、the acceptance8criteria, then the strict objective performance requirements of this section are not met.92.1.1 Strict Encoder/Decoder Objective Test Definition10The strict encoder/decoder objective test is intended to validate the implementation of the speech codec under11test. The strict objective
21、test is based upon statistical comparisons between the output of the master codec (see122.1) and combinations of the test encoder and decoder as listed below:13 Master encoder/test decoder (Stage 1)14 Master encoder with impaired packets/test decoder (Stage 2)15 Test encoder/master decoder (Stage 4)
22、16 Test encoder/test decoder (also Stage 4)17This test is a measure of the accuracy of the codec being tested in terms of its ability to generate a bit-stream18identical (or nearly identical) to that of the master codec when supplied with the same source material. The test19is based upon statistics
23、of the signal-to-noise ratio per 5 ms time segment, SNRi. Defining the output samples of20the master codec as xi(n) and those of the combination under test as yi(n), the SNRi for a segment i is defined as2122SNRi = 10 log10 PxiRi for Pxi or Pyi = T0 for Pxi and Pyi 64 bin 647The procedure is outline
24、d below, where the function int(x) truncates to the nearest integer whose value is8less than or equal to the value of x:910snr1:n SNRi values, snri = 0.0 indicates silence11h1:64 histogram values12npoint number of data points used to compute histogram13SNRSEG average value of SNRi values14SNRSEG = 0
25、.015npoint = 016for (i = 1, N)17if(snri not equal to 0.0) then1819npoint = npoint + 120SNRSEG = SNRSEG + snri21if(snri 64.0) then24h64 = h64 + 125else26hint(snri) = hint(snri) + 12728SNRSEG = SNRSEG / npoint2930The histogram array is converted into a sample cumulative distribution function, f1:n, as
26、 follows:3132h1:64 histogram values33npoint number of data points used to compute histogram34f1:64 normalized cumulative distribution function35TIA/EIA/I125-A2-7f1 = h1/npoint1for (i = 2, 64)2fi = fi-1 + hi / npoint3An ANSI C source language program, cdf.c, that implements this procedure is given in
27、 3.3.3.42.1.3 Minimum Performance Requirements5The codec being tested meets minimum performance requirements if all of the three following requirements are6met over the entire source speech material database of directory /expt1/objective:7 f14 = 0.01, which means that the SNRi can have a value lower
28、 than 15 dB no more than 1% of the time;8and9 f24 = 0.05, which means that the SNRi can have a value lower than 25 dB no more than 5% of the time;10and11 SNRSEG = 40 dB.122.2 Combined Objective/Subjective Performance Requirements13If the strict objective test has failed or is not attempted, then the
29、 codec being tested may be evaluated with the14combined objective and subjective tests described in this section. These objective and subjective tests are15divided into three experiments:16 Experiment I tests the codec under various speaker and input level conditions. Both objective and17subjective
30、tests shall be met to satisfy minimum performance requirements.18 Experiment II tests the decoder under impaired channel conditions. If the objective performance criteria19of this test are met, it is not necessary to perform the subjective component of the test. If the objective20performance criteri
31、a of this test are not met, the subjective test for channel impairments shall be met to21satisfy minimum performance requirements.22 Experiment III of the test insures that the rate determination algorithm in the encoder performs correctly23under a variety of background noise conditions. If the obje
32、ctive performance criteria of this test are24satisfied, it is not necessary to perform the subjective component of the test. If the objective25performance criteria of this test are not met, and the average data rate of the codec being tested is less26than 1.05 times the master codec average data rat
33、e, the subjective test for background noise degradation27shall be met to satisfy minimum performance requirements. If the average data rate of the codec being28tested is greater than 1.05 times the master codec average data rate, then the codec being tested fails the29minimum performance requirement
34、s.30Figure 2.2-1 summarizes the steps necessary to satisfy the combined objective/subjective test requirements.31The objective performance requirement for Experiments I and II (see 2.2.1 and 2.2.2) evaluate the decoder32component of the codec being tested. The philosophy of the test is the same as t
35、he strict objective test (see332.1). The objective performance requirement for Experiment III evaluates the encoder component of the codec34being tested, specifically the rate determination mechanism. The method of measurement for comparing the rate35output from the master encoder and the test encod
36、er is described in 2.2.3.2.36The subjective performance requirements for Experiments I, II and III (see 2.2.4) evaluate the codec being tested37against the master codec performance through the use of a panel of listeners and a Mean Opinion Score criteria.38TIA/EIA/IS-125-A2-81TIA/EIA/I125-A2-9codec
37、passes objective EXPT II for impaired channel conditionscodec passesStartYESNOcodec failsNONOYESYESYESNONONONOcodec passes object ive EXPT I for all 3 levelscodec passes subjective EXPT I M OS testcodec passes subjective EXPT II MOS testIs avg rate less than 1.05 times master codec ?codec passes sub
38、jective EXPT III MOS testcodec passes objective EXPT III testing of rate deter mination algorit hmYESYESYES1Figure 2.2-1. Objective/Subjective Performance Test Flow2TIA/EIA/IS-125-A2-102.2.1 Objective Performance Experiment I12.2.1.1 Definition2The objective performance Experiment I is intended to v
39、alidate the implementation of the decoder section of the3speech codec being tested under various speaker and input level conditions. The objective performance4Experiment I that tests the decoder is based upon the same philosophy as the strict objective test (see 2.1.1).5Only one combination of encod
40、er and decoder is evaluated against the master encoder/decoder, with the master6encoder driving the test decoder. The performance requirements are based upon sample statistics SNRSEG and7f1:n, which estimate segmental signal-to-noise ratio, SNRSEG, and the cumulative distribution function,8F(x), of
41、the signal-to-noise ratios per segment, SNRi. The codec being tested passes the objective performance9requirement of Experiment I if the master encoder/test decoder combination satisfies the criteria of 2.2.1.3.102.2.1.2 Method of Measurement11The method of measurement for decoder comparison is the
42、same as for the strict objective test (see 2.1.2).12Note: the post-filter shall be used in these objective experiments.132.2.1.3 Minimum Performance Requirements for Experiment I14The codec being tested meets the minimum requirements of the objective performance Experiment I, if all of the15three cr
43、iteria are met over the entire source speech material database in directory /expt1/objective.16The objective test shall be run three times, once for each of the three levels, -29, -19, and -9 dBm0. These levels17are provided in the source files18/expt1/objective/expt1obj.x29,19/expt1/objective/expt1
44、obj.x19 and20/expt1/objective/expt1obj.x09.21The numeric bounds given in Sections 2.2.1.3.1 through 2.2.1.3.3 are the performance requirements for Experiment22I.232.2.1.3.1 Minimum Performance Requirements for Experiment I Level -9 dBm024 f1 = 0.01, which means that the SNR can have a value lower th
45、an 2 dB no more than 1% of the time; and25 f15 = 0.06, which means that the SNR can have a value lower than 16 dB no more than 6% of the time;26and27 SNRSEG = 22 dB.282.2.1.3.2 Minimum Performance Requirements for Experiment I Level -19 dBm029 f1 = 0.01, which means that the SNR can have a value low
46、er than 2 dB no more than 1% of the time; and30 f15 = 0.14, which means that the SNR can have a value lower than 16 dB no more than 14% of the time;31and32 SNRSEG = 20 dB.33TIA/EIA/I125-A2-112.2.1.3.3 Minimum Performance Requirements for Experiment I Level -29 dBm01 f1 = 0.03, which means that the S
47、NR can have a value lower than 2 dB no more than 3% of the time; and2 f15 = 0.37, which means that the SNR can have a value lower than 16 dB no more than 37% of the time;3and4 SNRSEG = 16 dB.52.2.2 Objective Performance Experiment II62.2.2.1 Definition7The objective performance Experiment II of the
48、combined test, which validates the implementation of the decoder8section of the speech codec under impaired channel conditions, is based on the same philosophy as the strict9objective test (see 2.1.1). Only one combination of encoder and decoder is evaluated against the master10encoder/decoder: the
49、master encoder driving the test decoder.11Impaired channel packets from the master encoder used to drive the master decoder and test decoder for this test12are found in the file /expt2/objective/expt2obj.pkt of the companion software. The performance13requirements are based upon sample statistics SNRSEG and f1:n, which estimate segmental signal-to-noise14ratio, SNRSEG, and the cumulative distribution function, F(x), of the signal-to-noise ratios per segment, SNRi.15The codec under test passes, if t
