hash stringlengths 32 32 | doc_id stringlengths 7 13 | section stringlengths 3 121 | content stringlengths 0 2.2M |
|---|---|---|---|
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 2.1 Normative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the reference document (including any amendments) applies. Referenced documents which ar... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 2.2 Informative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the reference document (including any amendments) applies. NOTE: While any hyperlinks in... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 3 Abbreviations | For the purposes of the present document, the following abbreviations apply: ALG Application Layer Gateway ATS Abstract Test Suite B4 (NAT64) Basic Bridging BroadBand element ETSI ETSI TS 103 238-2 V1.1.1 (2015-01) 6 CPE Customer Premises Equipment DF Do not Fragment flag (in IPv4 header) FTP File Transfer Protocol GRT... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 4 Test Suite Structure | The identifier of the TP is built according to table 1 as recommended in the MTS methodologies. Table 1: TP naming convention |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5 Test Purposes | Proposes a TP proforma which is used in the present document. The fields of this proforma as used in the present document are explained in table 2. TP/<root>/<gr>/<sgr>/<x>/<nn> <root> = root NAT64 Mapping of Address and Port – Encapsulation Mode <gr> = group LSN Large Scale NAT <sgr> = sub-group BF Basic Function NP N... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1 TPs for LSN | |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.1 Basic Function | TP Id TP/NAT64/LSN/BF/BV/01 Test objective Check that the IUT supports the functionality of NAT64 1:1 NAT mapping Reference [1]:6.5.7.9 1:1 IP Mapping Initial conditions with { the IUT being properly provisioned and the interfaces are connected & functional } Expected behaviour ensure that { when { the IUT receives mul... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.2 NAT Pools | TP Id TP/NAT64/LSN/NP/BV/01 Test objective Check that the IUT supports the functionality of multiple NAT pools per prefix Reference [1]:6.4.6.10 Feature: NAT Grouping resource Sharing // [1]:6.4.4.4] Feature: Shared/Split Resources Initial conditions with { the IUT was properly provisioned the interfaces are connected ... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.3 Address Withdrawal | P Id TP/NAT64/LSN/AW/BV/01 Test objective Check that the IUT supports LSN GW address withdrawal on cache failure Reference [1]:6.4.6.16 NAT64 Address Withdrawal Initial conditions with { the IUT was properly provisioned the interfaces are connected & functional } Expected behaviour ensure that { when { the IUT receives... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.4 Fragmentation | TP Id TP/NAT64/LSN/FRAG/BV/01 Test objective Check that the IUT fragments an HTML IPv4 packet downstream Reference [1]:6.4.6.20 LSN Fragmentation and Buffering Initial conditions with { the IUT was properly provisioned the interfaces are connected & functional the physical MTU (Phy-MTU) size being equal or greater than... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.5 MSS Clamping | TP Id TP/NAT64/LSN/MSSC/BV/01 Test objective Check that the IUT functions with MSS clamping Reference [1]:6.4.9 Technical Viability Initial conditions with { the physical MTU (Phy-MTU) size being equal or greater than the IPv6 packet between all devices and the MTU (IPv6-MTU) being lower than the originating IPv6 packe... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.6 Static Port & IP Reservation | TP Id TP/NAT64/LSN/SPR/BV/01 Test objective Check that the IUT functions with static port reservation per prefix downstream Reference [1]:6.4.6.22 Feature: Port Reservation Initial conditions with { the IUT was properly provisioned the interfaces are connected & functional the static entries of well known ports for a s... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.7 NAT Timers | TP Id TP/NAT64/LSN/NT/BV/01 Test objective Check that the IUT TCP_time_wait timer expires when required Reference [1]:6.4.6.4 NAT64 LSN timers Initial conditions with { the IUT being properly provisioned and the interfaces are connected & functional and the IUT TCP_time_wait timer being set and the IUT having received ... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.8 Application Layer Gateway | TP Id TP/NAT64/LSN/ALG/BV/01 Test objective Check that the IUT supports FTP forwarding through an ALG Reference [1]:6.4.1 LSN Feature Summary Initial conditions with { the IUT being properly provisioned and the interfaces are connected & functional and the IUT being configured with FTP ALG set to active and the FTP cli... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.9 Routing Tables | TP Id TP/NAT64/LSN/RT/BV/01 Test objective Check that the IUT supports forwarding from GRT TO VRF Reference [1]:6.4.1 LSN Feature Summary Initial conditions with { the IUT being properly provisioned, and the interfaces are connected & functional, and the routing tables are configured GRT upstream ingress & VRF upstream... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.10 Anycast Addressing | TP Id TP/NAT64/LSN/AA/BV/01 Test objective Check that the IUT supports Anycast GW addressing Reference [1]:6.4.6.14 Feature: Anycast Gateway Address Initial conditions with { the IUT is properly provisioned the interfaces are connected & functional the IUT is configured with an Anycast address ] Expected behaviour ensu... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.11 Redundancy | TP Id TP/NAT64/LSN/RED/BV/01 Test objective Check that the IUT supports Redundant NPUs Reference [1]:6.4.1 NAT64 LSN Technology Feature Summary Initial conditions with { the IUT is properly provisioned the interfaces are connected & functional the IUT is configured with redundant NPUs ] Expected behaviour ensure that {... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.12 Load-Balancing | TP Id TP/NAT64/LSN/LB/BV/01 Test objective Check that the IUT supports load-balancing across NPUs Reference [1]:6.4.4.5 Feature: Traffic Based Load Balanced NPUs Initial conditions with { the IUT is properly provisioned the interfaces are connected & functional the IUT is configured with multiple NPUs ] Expected behavi... |
443c8f283ffc488872ea282fccb3278f | 103 238-2 | 5.1.13 Failure Events | TP Id TP/NAT64/LSN/FE/BV/01 Test objective Check that the IUT withdraws its IPv6 prefix when all NPUs fail Reference [1]:6.4.6.16 Feature: NAT64 Address Withdrawal Initial conditions with { the IUT is properly provisioned the interfaces are connected & functional the IUT is configured with multiple NPUs ] Expected beha... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 1 Scope | The present document describes testing methodologies which can be used to objectively evaluate the performance of super-wideband and fullband mobile terminals for speech communication in the presence of background noise. Background noise is a problem in mostly all situations and conditions and needs to be taken into ac... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 2 References | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 2.1 Normative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the referenced document (including any amendments) applies. Referenced documents which a... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 2.2 Informative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the referenced document (including any amendments) applies. NOTE: While any hyperlinks i... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 3 Definition of terms, symbols and abbreviations | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 3.1 Terms | Void. |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 3.2 Symbols | Void. |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 3.3 Abbreviations | For the purposes of the present document, the following abbreviations apply: AMR Adaptive Multi-Rate (narrowband speech codec) AMR-WB Adaptive Multi-Rate Wideband (wideband speech codec) AS Analysis Serial ASL Active Speech Level BAK Background Noise Component BGN Background Noise BM Basilar Membrane CM Cochlear Model ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 4 Introduction | The present document describes models for the objective prediction of speech-, background-noise- and overall quality for super-wideband and fullband terminals and systems used in background noise in uplink on a fullband scale. The models are intended to be used for modern terminals including e.g. different bitrates of ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 5 Underlying speech databases and preparations | The basis of any perceptually-based measure which models the behaviour of human test persons, are auditory tests. In general, these tests are carried out with naïve test persons, who are asked to rate a certain quality aspect of a presented speech sample. For the evaluation of processed and transmitted noisy speech, th... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6 Model descriptions | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.1 Introduction | The prediction models described in the following clauses are full-reference models. Such a predictor compares the degraded signal under test against a reference signal. Audible disturbances between these two signals are assumed to highly correlate with the results of auditory tests conducted in the development phase. T... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.2 Common definitions | Even though both model variants internally work differently regarding the processing steps, inputs and outputs are common. The input time signals are denoted as x(k) for the reference signal and y(k) for the degraded signal, which is evaluated either by the instrumental or the auditory assessment. Each prediction model... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3 Model A | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.1 Introduction | In general, the model consists of several stages and calculation steps which finally conclude in the assessment of instrumental S-, N- and G-MOS. Figure 6.1 provides an overview about the structure of the method. Clauses 6.3.2 to 6.3.8 provide detailed descriptions of each processing block. Figure 6.1: Block diagram of... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.2 Pre-Processing | The pre-processing of the inputs x(k) and y(k) is conducted to compensate differences regarding temporal alignment and level offsets between the signals. An overview of the pre-processing is given in the block diagram shown in figure 6.2. Figure 6.2: Block diagram of model A The first block ensures that both input sign... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.3 Spectral transformation | Figure 6.3 depicts the flow chart of the spectral transformation and further processing steps, which are performed for the instrumental assessment of S-MOS, N-MOS and G-MOS. Figure 6.3: Block diagram of model A For the consideration of a hearing-adequate signal representation, the model according to [i.20] and [i.21] i... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.4 Non-linear loudness transformation | The determined spectral representations are transformed with a hearing-adequate filterbank, but are still provided in the physical unit Pascal. The real loudness as perceived by humans is not yet considered. For this purpose, the aforementioned hearing model introduces a non-linear transformation to address the compres... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.5 Instrumental assessment of N-MOS | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.5.1 Introduction | The flow chart for the instrumental assessment of N-MOS is depicted in figure 6.6. Several metrics are derived from the spectral representation and then mapped versus the auditory BAK results. Only the separated noise estimation , is evaluated for the determination of these metrics. Figure 6.6: Instrumental assessm... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.5.2 Loudness-based features | The first step of the calculation is to apply an A-weighting function [i.23] to the noise-only spectrum (output of spectral transformation stage), which results in a weighted version , . Both, weighted and unweighted noise spectra are then transformed according to the original non-linear transformation of sound pre... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.5.3 Sharpness-based feature | The previously introduced metrics do not explicitly consider the higher frequency range, which is even more attenuated by the A-weighting function. For this purpose, the unweighted version of the transformed noise spectrum is used for a sharpness metric. This well-known psycho-acoustic measure especially addresses high... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.6 Reference optimization and asymmetry | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.6.1 Introduction | A common processing step in speech quality prediction algorithms (e.g. [i.24] or [i.25]) is the so-called reference optimization. When comparing spectral representations of a degraded and a reference signal e.g. by a subtraction, often differences occur which are not perceived by test subjects. The aim of this optimiza... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.6.2 Reference optimization | The reference optimization compensates inaudible signal modifications between reference and degraded. All optimizations are only conducted on the reference. This processing is divided into three parts: • Compensation of the group delay: differences in group delays between reference and degraded signal are almost inaudi... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.6.3 Masking of inaudible differences | So far, reference optimization only considered effects of the signal processing of the degraded signal which are inaudible to human listeners. Beside this, there are also psycho-acoustic masking effects which can impact the audible differences between degraded and reference signal. In this stage, the masking model desc... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.6.4 Asymmetry | The final block of this stage is the application of the so-called "asymmetry". This principle was already introduced in several instrumental speech quality assessment methods, like e.g. [i.29], [i.25] or [i.24]. In auditory tests, additional signal components are much more annoying than components which are removed. A ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7 Instrumental assessment of S-MOS | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7.1 Introduction | 6.3. |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 7.1 Introduction | For the instrumental assessment of S-MOS, two different kinds of features are extracted from the spectral representations. The first type of metrics is intended to detect disturbances at certain modulation frequencies and detects degradations versus frequency. The second one is intended to determine spectral difference... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7.2 Modulation-based features | Figure 6.10 provides a flow chart of the calculation of the two modulation-based features. The spectral representations of the separated speech , and the pre-processed, but non-optimized reference 3$, are used as the inputs of this block. Figure 6.10: Structure of modulation-based feature extraction In a fi... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7.3 Spectral difference features | A more classic approach for the instrumental assessment of speech degradation is the evaluation of spectral differences of loudness. The differences of the non-linearly compressed sound pressure level are used here. Figure 6.11 illustrates the flow chart for the determination of several features. ETSI ETSI TS 103 281 V... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7.4 Control parameters | In addition to the psycho-acoustically motivated metrics of the previous clauses and the aforementioned SNR(A), two so-called control parameters are introduced. These single values do not necessarily correlate directly with perceived speech quality, but provide a weighting for the other metrics. The equivalent rectangu... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.7.5 Combination of features | As for the N-MOS determination, the derived features again are combined by a random forest regression model. Table 6.4 provides the parameters for the regressor used for the training. Figure 6.12 illustrates the inputs which are found to be crucial for the instrumental assessment of speech distortion component (SIG). T... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.3.8 Instrumental assessment of G-MOS | The instrumental assessment of G-MOS is based on three input features. Similar to the prediction model in [i.2], the previously determined values of S-MOS and N-MOS are used here as well. Obviously, this usage is close to the judgement of test subjects in an auditory test. Several investigations already indicated that ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4 Model B | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.1 Overview | The SWB/FB Cochlear model (CM) based Prediction algorithm compares a reference signal x(k) with a signal under test y(k), and estimates the results of a listening-only test on the three speech quality attributes. The three speech quality attributes and type of listening test estimated are according to Recommendation IT... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.2 Operational Modes | The model algorithm supports a single fullband operational mode and can predict scores for narrowband, wideband, super-wideband and fullband degraded speech signals. All the test conditions are re-sampled to 48 kHz prior to input into the Auditory Model. Regardless of the bandwidth of signals under test, the model algo... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.3 Temporal Alignment | The Temporal Alignment module is currently using a subsample alignment algorithm [i.33]. This algorithm detects the delay between original and degraded wave speech by interpolation of the peak of cross correlation. A parabola curve is used for the three-point peak interpolation. The method has been verified to perform ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.4 Voice Activity Detection (VAD) and segment classification | The VAD module classifies speech into voiced/unvoiced/background (noise only) sections that will contribute to different distortions. The background (with background noise only) section will affect background dimensions (N-MOS). The voiced sections are the main part that decide the speech quality dimensions (S-MOS). Th... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5 Auditory Model | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.1 Introduction | Speech is converted into perceptual domain with the help of a Cochlear Model. The purpose of the Cochlear Model is to compute the Inner Hair Cell (IHC) response caused by sound impinging at the input to the Ear Canal. The model can be conceptually divided into five parts: Ear Canal model, Middle Ear model, Hydro-mechan... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.2 Ear Canal model | The ear canal model is a linear model with a primary resonance at 2 600 Hz. It models the transfer function between pressure incident at the Ear Canal opening to the pressure on the Tympanic membrane (see figure 6.15). As such, it is essentially an acoustic model of a single ended cylindrical tube [i.31]. The length of... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.3 Middle Ear model | The middle ear model emulates the transfer function between pressure incident at the Tympanic membrane and the Volume Velocity of the Stapes (see figure 6.15). The model can be viewed as a transformer in series with the impedance of the Middle Ear and the input impedance of the Cochlea. The Middle Ear impedance is give... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.4 Hydro-mechanical cochlear model | The cochlear model (CM) is a spatially two-dimensional hydro-mechanical model, which computes various electrical and mechanical responses in the cochlea. In particular, the model can be used to calculate Basilar Membrane (BM) and Inner Hair Cell (IHC) response as a function of time and space. Detailed aspects of the co... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.5 Hair Cell transduction model | As mentioned above, the OHC and IHC outputs are derived from the rectified and low-pass filtered cilia displacement. This is shown in figure 6.17 and figure 6.18. The IHC and OHC models are thus alike except for a high-pass filter that precedes the IHC model to account for the fact that the IHC cilia are not attached t... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.5.6 Outer Hair motility model | The cochlear non-linearity imposed by OHC motility is modelled as mechanical feedback from the OHC, which modifies the macro-mechanical impedance. This is shown in figure 6.19 as "Slow Acting Active Feedback". Figure 6.19: OHC motility |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.6 Feature Extraction | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.6.1 Introduction | Perceptual features are extracted from the CM output before mapped to objective prediction scores. There are two main features that are used. The Salient Formant Points (SFP) are used for time-related distortion measurement. The Cochlear Output Statistical Metric (COSM) are used for frequency related distortion (S-MOS)... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.6.2 Salient Formant Points (SFP) feature extraction | The aim of the SFP feature set is to isolate and predict the temporally localized distortions such as single 'clicks' and 'pops', but also more temporally dense distortions which produce the perception of 'harshness'. Generally speaking, and in line with Principal Component Analysis (PCA), the distortions can be classi... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.6.3 COSM (Cochlear Output Statistic Metric) feature extraction | The COSM feature set is extracted from: 1) voiced sections, which are salient for determining foreground distortions (S-MOS); and 2) background sections (where only background noise exists) for determining background distortions (N-MOS). The COSM feature extraction operates on the output of the physiological Cochlear M... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.4.7 Training and mapping | The SFP and COSM features sets are converted into the mean opinion objective quality scores through a mapping procedure. An a-priori training process based on deep neural network (DNN) is used to determine suitable weighting factors for each of the features. The a-priori training is based on a large set of listening te... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 6.5 Mapping of model outputs | As shown in annex A, the prediction models were trained with a large number of databases including a wide range of qualities. Even though reference conditions according to [i.17] were used in all databases, each subjective experiment has its own context, i.e. a certain average, minimum or maximum quality. This effect c... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 7.2 Results for Model A | Table 7.1: Training results for model A after 3rd order mapping Feature vs. Database S-MOS/SIG N-MOS/BAK G-MOS/OVRL DES-01 DES-02 DES-03 ETSI ETSI TS 103 281 V1.3.1 (2019-05) 37 Feature vs. Database S-MOS/SIG N-MOS/BAK G-MOS/OVRL DES-04 DES-05 DES-06 DES-07 ETSI ETSI TS 103 281 V1.3.1 (2019-05) 38 Feature vs. Database ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 7.3 Results for Cochlear Prediction Model (Model B) | Table 7.2: Training results for model B after 3rd order mapping Feature vs. Database S-MOS/SIG N-MOS/BAK G-MOS/OVRL DES-01 DES-02 ETSI ETSI TS 103 281 V1.3.1 (2019-05) 42 Feature vs. Database S-MOS/SIG N-MOS/BAK G-MOS/OVRL DES-04 DES-05 DES-06 DES-07 ETSI ETSI TS 103 281 V1.3.1 (2019-05) 43 Feature vs. Database S-MOS/S... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8 Validation results | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.1 Introduction | For the validation of the models different databases were provided. The databases included different types of conditions of different terminals and simulations. The details of the validation databases are described separately for each validation database. As validation database 1 was also one of the training databases ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.2 Validation database 1 (DES-17) | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.2.1 Database description | Validation database 1 was created using an acoustic mockup with multiple microphones mounted in locations typical of contemporary handsets. The acoustic set-up was consistent with the handset and hand-held handsfree conditions described in ETSI TS 126 132 [i.14], with the acoustic mockup mounted on, and in front of, HA... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.2.2 Validation database 1: Results for model B | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model B on validation database 1 for each of the three ratings, SIG, BAK, and OVRL are shown in figure 8.1. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (rig... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.3 Validation database 2 (DES-20) | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.3.1 Database description | Validation database 2 was created using three commercial and pre-commercial handsets. Each was operated in handset and hand-held handsfree mode, set up as described in ETSI TS 126 132 [i.14]. The speech source was full-band Mandarin phonetically-balanced sentences, with two sentences from each of four male and four fem... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.3.2 Validation database 2: Results for model A | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 2 for each of the three ratings, SIG, BAK and OVRL, are shown in figure 8.2. As in figure 8.1, for each rating (rows), two scatter plots are shown, one before a monotonic mappi... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.4 Validation database 3 (DES-25) | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.4.1 Database description | Validation database 3 was created using two commercial handsets. Each was operated in handset, headset, and hand-held handsfree mode, set up as described in ETSI TS 126 132 [i.14]. The speech source was full-band American English phonetically-balanced sentences, with two sentences from each of four male and four female... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.4.2 Validation database 3: Results for model A | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 3 for each of the three ratings, SIG, BAK, and OVRL, are shown in figure 8.3. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (ri... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.4.3 Validation database 3: Results for model B | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 3 for each of the three ratings, SIG, BAK and OVRL, are shown in figure 8.4. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (rig... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.5 Validation database 4 (DES-26) | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.5.1 Database description | Validation database 4 was created using two commercial handsets, different from those of database 3. Each was operated in handset, headset, and hand-held handsfree mode, set up as described in ETSI TS 126 132 [i.14]. The speech source was full-band American English phonetically-balanced sentences, with two sentences fr... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.5.2 Validation database 4: Results for model A | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 4 for each of the three ratings, SIG, BAK, and OVRL, are shown in figure 8.5. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (ri... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.5.3 Validation database 4: Results for model B | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 4 for each of the three ratings, SIG, BAK and OVRL, are shown in figure 8.6. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (rig... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.6 Validation database 5 (DES-27) | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.6.1 Database description | Validation database 5 was created using three commercial handsets, different from those of databases 3 and 4. Two were operated in handset, headset, and hand-held handsfree mode, while one was operated in handset and hand-held speakerphone modes. Set up was as described in ETSI TS 126 132 [i.14]. The speech source was ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.6.2 Validation database 5: Results for model A | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 4 for each of the three ratings, SIG, BAK and OVRL, are shown in figure 8.8. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (rig... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 8.6.3 Validation database 5: Results for model B | Results are shown as scatter plots, comparing instrumental predicted ratings to subjective ratings. Results from model A on validation database 4 for each of the three ratings, SIG, BAK and OVRL, are shown in figure 8.9. For each rating (rows), two scatter plots are shown, one before a monotonic mapping is applied (rig... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9 Application of the models | |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9.1 Introduction | In order to avoid ambiguities in the results, the objective model should be applied in the way it was applied during the training process which also reflects the listening test. |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9.2 Speech material | The speech samples used in conjunction with the model should be the ones used in the subjective tests of the training databases. At least 16 sentences of male and female speakers shall be used per measurement run. Each sentence shall be centered in a time window of 4,0 seconds. The minimum duration of an active speech ... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9.3 Positioning of the device under test | For testing in the handset use case, variations from nominal position defined in clause 8.4 of Recommendation ITU-T P.64 [i.18] representing typical use conditions can also be used. For testing headsets, positioning recommendations found in Recommendation ITU-T P.380 [i.40] should be followed. For testing handheld hand... |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9.4 Background noise playback | The setups according to ETSI ES 202 396-1 [i.1] and ETSI TS 103 224 [i.19] are recommended for the generation of the background noise. The background noises to use in conjunction with the model shall be taken from these two standards. |
7e3fac1755250572ce2c8dd45a4c4086 | 103 281 | 9.5 Recording and calibration procedure | In general, the signal to be evaluated with the models is recorded at an electrical reference interface. Typically the signal is obtained in a certain physical unit (like Volt) or on a digital scale (e.g. 16-bit integer representation). The level calibration to the acoustical domain (in dB Pa or dB SPL) according to th... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.