TECHNICAL WHITE PAPER. Audio Loudness Analysis

Transcription

1 TECHNICAL WHITE PAPER Audio Loudness Analysis Samuel Fleischhacker, March 2014

2 INTRODUCTION LOUDNESS MEASUREMENT FUNDAMENTALS Overview Loudness measurement Loudness range measurement True peak level measurement CMA 1820 COMPRESSED MEDIA ANALYZER Summary Loudness analysis...13 CONCLUSION Page 2

3 INTRODUCTION Audio loudness jumps when changing TV channels or programs or on ad insertion are a major source of irritation for TV viewers. Many complaints have been received by the regulatory authorities (FCC in the United States, CSA in France, AGCOM in Italy, Ofcom in the UK, etc.). These authorities are therefore aware of the problems and governments and regulatory bodies worldwide are forcing broadcasters to comply with audio normalization standards. Regardless of any legal requirement, it is in broadcasters best interests to harmonize audio levels in their system in order to improve the viewer s experience. As an initial solution, predictable and well-characterized methods have been defined and standardized to accurately measure audio loudness irrespective of content (news, sport, advertisements, music, drama or movies). The first part of this paper introduces the audio loudness measurement fundamentals. The second part focuses on the measurement analysis tool available from Sencore, namely the CMA 1820 Compressed Media Analyzer. 1. LOUDNESS MEASUREMENT FUNDAMENTALS 1.1 Overview Audio loudness analysis is based on the following three key measurements: Program Loudness Defined by the International Telecommunication Union in ITU-R BS.1770, loudness measures how loud a program is on average. The result is a value expressed in either LKFS (Loudness K weighted Full Scale) or LUFS (Loudness Unit Full Scale). These are simply alternative names for the unit but the measurement and result are the same. The recommended program loudness is -23 or -24 LUFS. Loudness Range (LRA) Defined by the European Broadcasting Union in EBU Tech Doc 3342, the Loudness Range (LRA) measures loudness variation within a program (range between quiet and loud sounds). Compression is recommended in the case of a high LRA value. The result is a value expressed in LU (Loudness Unit). The recommended LRA is 20 LU to achieve a comfortable listening level. Maximum True Peak Level (TPL) Defined by the International Telecommunication Union in ITU-R BS.1770, the true peak level indicates the maximum value of the audio signal waveform. The true peak should not exceed a recommended level to avoid any audio saturation or clipping. The result is expressed in dbtp (decibel True Peak). The recommended value is -1 dbtp (which means the peak is 1 db below a digital full scale signal). All these measurements are described later in this paper. Page 3

4 Apart from the ITU-BS.1770 standard, several recommendations have been adopted in various regions around the world. These documents provide explanations and advice about target loudness measurements. Some examples: In Europe, the European Broadcasting Union (EBU) P/LOUD group developed the following recommendations based on ITU-R BS : Recommendation R128, Tech 3341, 3342, 3343 and In the US, the Advanced Television Systems Committee (ATSC) released recommended practice A/85 to comply with the CALM Act. In Japan, the Association of Radio Industries and Businesses (ARIB) developed the TR-B32 standards. In Australia, the operational practice OP-59 was developed by FreeTV. Figure 1 Loudness standard overview Page 4

5 1.2 Loudness measurement Audio Loudness measurement is described in ITU-R BS Loudness refers to the perceived strength of a piece of audio (music, speech, sound effects, etc.) and depends on the audio level, frequency, content type and duration. The measurement is expressed in LKFS (ITU-R BS.1770) or LUFS (EBU Tech Doc 3341). LKFS and LUFS are strictly equivalent. How is loudness measured? Loudness can be measured on mono, stereo and 5.1 audio signals. In the case of 5.1 audio signals, the five main channels (left, center, right, left surround and right surround) are considered. The LFE (low frequency effects) channel is excluded from the measurement. The following figure describes how loudness is calculated step by step. K weighting and filtering Mean square calculation Sum of weighted channels Gating Figure 2 Loudness measurement algorithm Page 5

6 K weighting and filtering is a pre-filtering stage as illustrated below. Filtering is applied to all audio channels (except LFE) and is designed to model the acoustic effect of the head. The ITU stated that high frequencies are perceived as louder (and more irritating) than normal frequencies. This is why the weighting curve is modeled to: Attenuate the low frequencies (< 100 Hz) Preserve the medium frequencies (between 100 Hz and 1000 Hz) Amplify (+4 db) the high frequencies (> 1000 Hz) Figure 3 K-weighting and filtering The Mean square calculation calculates the power of pre-filtered samples within an interval (T) of 400 milliseconds. This operation is applied to each audio channel (except LFE). Considering a 48 khz sampled audio signal, 400 ms corresponds to 19,200 samples per channel. The mean square for the left channel is therefore L. where Lx = xth sample for the left channel The same applies for the other channels (center, right, left surround and right surround). The channel powers for the sum of weighted channels are as follows. The left/right surround channels are amplified (+ 1.5 db x1.41) whereas the left/right/center channels are not (0 db x1). The result is the momentary loudness LK LK = log10 (L + R + C x LSurround x RSurround) Page 6

7 Gating serves to pause the loudness measurement when the signal drops below a given threshold. The step is performed by considering all momentary loudness samples (third step) from consecutive and overlapping intervals (T) of 400 milliseconds. The overlap should be 75% of the interval. A two-stage process is then used to perform the gated measurement. With an absolute threshold of -70 LKFS With a relative threshold of -10 LU below the average of the measurements made after the absolute threshold Figure 4 Gating Momentary loudness (M) is illustrated in the figure above. It corresponds to loudness using a 400 ms sliding window with no gating. Short-term loudness (S) is not illustrated in the figure above. It corresponds to loudness using a 3 s sliding window with no gating. It is basically equivalent to momentary loudness but with a long measurement period. Integrated loudness (I), also known as program loudness, is illustrated in the figure above. It uses the gating function (absolute and relative). Page 7

8 1.3 Loudness range measurement The Loudness range (LRA) measurement is described in EBU Tech Doc The Loudness range (LRA) quantifies the loudness variation within a program. It is a generic measurement (in LU) that helps to decide whether dynamic compression is required. There is no set LRA target value but initial experiments suggest a maximum LRA value of 20.0 LU for highly dynamic content, such as action movies or classical music. Other content can have a lower LRA value. How is the loudness range measured? The LRA is based on a statistical distribution of loudness, using a 3 s sliding window for integration. Loudness distribution is illustrated in the figure below. Figure 5 Loudness distribution In this distribution, all loudness measurements below the following thresholds are excluded: -70 LUFS (absolute threshold). No relevant signals are generally found in this range. -20 LU relative to program loudness (relative signal dependent threshold). Background noise and periods of silence are removed. In our example with program loudness at LUFS, the relative threshold is LUFS. Page 8

9 Figure 6 Modified Loudness distribution Page 9

10 The LRA is the difference between the estimates of the 10th and 95th percentiles of the distribution. The lower percentile of 10% corresponds to a music fade-out. The upper percentile of 95% corresponds to a single unusually loud sound, such as a gunshot. In our example, the LRA is 25.0 LU. Figure 7 Loudness Range (LRA) Page 10

11 1.4 True peak level measurement The true peak level measurement is described in ITU-R BS The true peak level indicates the maximum (positive or negative) value of the signal waveform in the continuous time domain. The true peak should not exceed a recommended level to avoid any audio saturation or clipping. The maximum permitted true peak level during production should be -1 dbtp according to EBU and -2 dbtp according to ATSC. As previously mentioned, measurements of true peak level use the unit dbtp and are referenced to the digital full scale value 0 dbfs. 0 dbfs corresponds to a sample with the maximum level that can be represented. In a 16-bit digital two s complement format, this means (in binary) or 0x7FFF (in hexadecimal). If the highest peak sample is -2 db below 0 dbfs, this means that the true peak level is -2 dbtp. How is the true peak level measured? The true peak level indicates the maximum (positive or negative) value of the signal waveform in the continuous time domain. The value can be higher than the upper sample value in the 48 khz time-sampled domain. Figure 8 True Peak Level As illustrated in the figure above, the sample peak is not necessarily the real true peak. For this reason, the true peak measurement is achieved by 4 x oversampling the initial sampling rate (from 48 khz to 192 khz). Page 11

12 2. CMA 1820 COMPRESSED MEDIA ANALYZER 2.1 Summary The CMA 1820 Compressed Media Analyzer is a file-based media analyzer. It performs in-depth analysis of video and associated audio elementary streams, making it the perfect solution for troubleshooting, investigation, validation, debugging, interoperability testing and performance evaluation. The analyzer supports present and future generations of compressed audio/video formats, including: Video: H.265/HEVC, H.264/AVC, MPEG-2 and VC-1 (SD, HD and UHD) Audio: MPEG Audio, AC-3, E-AC3 and AAC (mono, stereo and surround) Figure 9 HEVC/AAC (stereo) analysis Sencore recently introduced audio loudness analysis capability for TV/Radio programs irrespective of the compression format. Page 12

13 2.2 Loudness analysis The CMA 1820 offers all the measurements previously described in this paper: Loudness (program/integrated, short-term and momentary) Loudness Range True Peak Figure 10 CMA 1820 Compressed Media Analyzer Loudness measurements The loudness analysis period is user-configurable. The selection to be analyzed can be determined by the beginning and end of the file, an audio frame range or a video picture range. Page 13

14 The tool checks the measurements are within the authorized/expected range. As shown in the following screenshot, the user can set the expected loudness measurement as either a target value, a ± tolerance or min/max values. Figure 11 Loudness value settings The results are displayed as numerical values with the following dedicated symbols Value in the expected range Value outside the expected range (too high) Value outside the expected range (too low) For momentary and short-term loudness, the CMA 1820 indicates the maximum and minimum measured values. It also counts the number of overflows (i.e. where loudness is above the max. threshold) and underflows (i.e. where loudness is below the min. threshold). in time-line graphs Time-line graphs help to navigate within the stream and locate when the measurements are out of range. Page 14

15 Figure 12 Loudness graph (Momentary Loudness) Finally, the CMA 1820 can export all the measurements as text and CSV files for post-analysis with spreadsheet applications. Figure 13 Loudness log export (text format) Page 15

16 CONCLUSION Thanks to predictable and well-characterized methods for measuring audio loudness, as well as appropriate analysis tools such as the CMA 1820, broadcasters can measure and investigate loudness issues for the benefit of TV viewers to ensure that there are no more level jumps when switching between TV programs or during commercial breaks. Copyright 2014 Sencore, Inc. All rights reserved Sencore Drive, Sioux Falls, SD USA This publication contains confidential, proprietary, and trade secret information. No part of this document may be copied, photocopied, reproduced, translated, or reduced to any machine-readable or electronic format without prior written permission from Sencore. Information in this document is subject to change without notice and Sencore Inc. assumes no responsibility or liability for any errors or inaccuracies. Sencore, Sencore Inc, and the Sencore logo are trademarks or registered trademarks in the United States and other countries. All other products or services mentioned in this document are identified by the trademarks, service marks, or product names as designated by the companies who market those products. Inquiries should be made directly to those companies. This document may also have links to third-party web pages that are beyond the control of Sencore. The presence of such links does not imply that Sencore endorses or recommends the content on those pages. Sencore acknowledges the use of third-party open source software and licenses in some Sencore products. This freely available source code can be obtained by contacting Sencore Inc. About Sencore Sencore is an engineering leader in the development of reliable, cost-effective signal transmission and content monitoring solutions for the broadcast, cable, satellite, and IPTV markets. Backed by world-class customer service and support, Sencore s portfolio includes video contribution and distribution equipment, system monitoring and analysis solutions, and test and measurement instruments. Designed to meet the needs of an ever-changing industry, Whether IP based, multiscreen OTT or readying for HEVC and 4k Sencore solutions leverage the latest technology to ensure efficient delivery of high-quality video from the source to the home. More information about Sencore can be found at All trademarks and registered trademarks mentioned herein are the property of their respective owners. Page 16