ABSTRACT

Loudness discrepancies have been a major source of headaches for the broadcast professional, and only seem to be getting worse in digital television broadcasts. Before corrective action can be taken, program loudness must be measured and the behavior of the professional integrated receiver/decoder (IRD) understood. A broadcaster's best efforts at controlling loudness and reducing the number and frequency of viewer complaints can be undermined by just a few outlier programs or a delivery path that has not been appropriately managed. Rather than approaching the task piecemeal, a systemic approach is necessary to achieve the best results and reduce viewer complaints.

WHY IS LOUDNESS SUCH A PROBLEM IN DIGITAL TELEVISION?

Current practices in NTSC broadcast television provision the audio signal such that it approaches the highest permissible level. These legacy practices have provided, as an unintended byproduct, a measure of program loudness normalization within a specific channel's broadcasts, as will be explained later in this paper. Inter-channel differences have been apparent (i.e. one channel may be louder or quieter than another channel), but inter-program loudness (i.e. different programs broadcast on a specific channel) has generally maintained a relatively consistent level due to these legacy practices.

Digital television has the capability of providing a cinematic experience to the viewer, complete with stunning visuals and dynamic, multichannel surround sound. Maintaining wide dynamic range audio within digital programming is a key and desirable feature of digital television, but wide dynamic range audio can present challenges to equipment, broadcasters and viewers alike.

Figure 1 shows three day's worth of the exact same primetime, high value programming presented and measured simultaneously on both the analog and the digital channels of a single terrestrial network affiliate. While the analog signals (the higher cluster of data points) clearly appear to be much louder than the digital signals (the lower, and more varied cluster of data points), the analog signals are clearly more consistent, program to program. Hence, when a viewer tunes to a specific channel, in many cases a single volume adjustment will be sufficient to "calibrate" their listening environment to their desired acoustic volume level, and most programs on that channel will not require a volume shift to maintain the viewer's desired target playback level. The digital signals (the lower and more varied collection of data points) show the wide dynamic range of high production value programming, and there is clearly a wide variance from program to program. Fortunately, systems and processes within most broadcasters' existing equipment and signal paths can help to maintain consistent levels within and between both analog and digital broadcast signals at no additional expense when understood and properly utilized.

UNDERSTANDING RECEIVER BEHAVIOR

At the local station or cable headend, the behavior of an integrated receiver/decoder is often overlooked, as long as downstream equipment appears to receive the appropriate signals and the appropriate program is on air with both sound and picture. Engineering resources are generally reserved for those systems that are reporting error conditions, and loudness complaints from viewers have historically not been accompanied by an error message displayed on encoding or processing equipment. Unbeknownst to many station and headend engineers is the fact that appropriate settings in the audio decoding subsystem within these IRDs can provide a major benefit to reducing viewer complaints and presenting an appropriate and compelling viewing experience in the home.

Digitally Tiered Analog Services

For those local stations and cable headends that are receiving a pre-encoded Dolby Digital (AC3) data stream and decoding this signal within the IRD (i.e. not using an external, professional reference decoder) for transmission on an analog tiered service or for other purposes (bug insertion, time shifting, etc.), inappropriate settings within the IRD audio decoding subsystems are a primary source of viewer complaints. The lack of understanding of these settings and the manufacturer's default conditions of the IRD are contributing to this issue. In order to best match existing analog programming on these delivery systems, a specific mode must be invoked within the Dolby Digital (AC3) decoding subsystem in order to maintain a match between programs and neighboring channels. This decoding mode is known as "RF Mode" in Dolby parlance, but IRD manufacturers may use another term such as "heavy" or "narrow" in their units. The behavior of this mode will be discussed in detail in the following section. It is important to understand that this mode will only affect the decoded analog or digital output of the IRD. Any selected decoding mode will not affect the Dolby Digital (AC3) bitstream output of the IRD, and those systems and signal paths conveying AC3 via a digital pass through architecture will maintain the capability of conveying full dynamic range audio to their viewers.

The Effects of Audio Metadata

Dolby Digital (AC3) data streams contain audio metadata, which provides necessary information to the receiver/decoder. A main component of this audio metadata is the dialog level parameter (aka dialnorm), which is intended to convey to the decoder the average A-weighted (or K-weighted) level of conversational speech within the audio program. Studies have shown that conversational speech and dialog elements within programming are the anchors by which viewers adjust their playback volume to a desired level. When the dialnorm parameter is set such that it accurately conveys the average A (or K)-weighted level of conversational speech, the average speech level in the decoded output of a consumer IRD will be maintained around -31dBFS. This provides enough headroom for a wide dynamic range, cinematic experience in the home via a digital television broadcast.

However, the dialog leveling subsystem via the dialnorm metadata parameter does not live in a vacuum. Other important processes within the AC3 decoding process depend on an accurately set dialnorm parameter. In addition to the most commonly understood function of provisioning the average level of speech to a specific target level and therefore "calibrating" the viewer's listening environment, the dialnorm parameter also provides the basis for dynamic range processing (aka dynamic range control, or DRC) within the Dolby Digital (AC3) decoder (Figure 2).

The dialnorm parameter, when properly reflecting the average level of conversational speech within the program, provides the center point to a "null band" where (when invoked manually by the viewer or as defined by the manufacturer to reflect the capabilities of the playback equipment) no DRC processing occurs. When used correctly, these two processes prevent inappropriate processing from occurring on conversational dialog within the program in addition to normalizing the level of speech to the target level of -31dBFS.

In the United States, cable systems typically receive content via both digital and analog methods, and relay this content often via both digital and analog systems to their subscribers. This combination of dual delivery methods requires the ability to provision digital services to match analog services with respect to level. Often, broadcast engineers are unaware that existing equipment likely contains all the necessary processing to match levels within their digitally tiered services without any additional expense, provided that the content itself is provisioned properly prior to delivery to the broadcaster. Methods and processes to properly provision content for delivery will be discussed later in this paper.

As most consumer digital IRDs are capable of receiving varied input signals (AC3, MPEG-1 Layer 2, QAM, DVB, ATSC, etc.) from multiple sources and provide the audio program within those received input signals at several different output connections, the gain structure within these IRDs is provisioned with the assumption that the dialnorm parameter within a received Dolby Digital (AC3) data stream is set correctly when tuned to a digital service. When tuned to an analog or other non-Dolby Digital service, the gain structure is provisioned with the assumption that the average level of conversational speech is about -17dB below 100% modulation.

Legacy NTSC signals are generally normalized via audio processing in an effort to maximize the permitted broadcast level. This has the effect of providing a "brute force" method of controlling loudness differences by minimizing dynamic range and making all audio elements (speech, music, sound effects, etc.) within a program near to the same level. The Dolby Digital (AC3) system contains specific processes that, when properly utilized, can help to match levels within digital programming to existing analog broadcasts while simultaneously helping to retain dynamic range to create a more transparent and compelling listening and viewing experience.

The decoded target level of speech at -31dB is too quiet when compared to standard NTSC broadcasts. In order to match the average level of speech and the wide dynamic range signals that are commonly found in digital broadcasts to existing analog programming, a specific dynamic range control (DRC) profile should be invoked to better match the narrow dynamic range of programs broadcast on analog services. When the dialnorm parameter accurately reflects the average A (or K)-weighted level of conversational speech in the encoded program stream, RF Mode DRC appropriately scales high level dynamics downwards and low level dynamics upwards thus "compressing" the dynamic range of the digital program (Figure 2). In addition, the Dolby Digital (AC3) decoder invokes an 11dB level shift upwards when RF Mode DRC is enabled to better match the expected level of speech within the gain structure of RF demodulators. When the dialnorm parameter accurately conveys the average level of conversational speech within the program, the 11dB shift above the normalized level of -31dB invoked by the RF Mode DRC setting places the average level of speech at -20dB for stereo programming, and when summed to mono, matches the expected target level within NTSC broadcasts of -17dB. In addition, RF Mode provides 6dB more headroom than typical NTSC programming, providing, "…a similar dynamic range to NTSC without the need for the violent non-linear processing commonly employed to avoid over-modulation."

The keys to the predictable behavior of a receiver and an appropriate and compelling audio experience in the home are the correct setting of these audio metadata parameters within the received program audio stream and the correct setting of the Dolby Digital (AC3) decoding subsystem within the professional IRD.

The Consequences of Improperly Set Audio Metadata Parameters within Delivered Signals

When audio metadata within a delivered Dolby Digital (AC3) data stream is correct, the licensed decoder within a receiver processes the audio prior to delivery to the output connections in an appropriate fashion based upon settings within the delivered audio metadata set. The main processes include leveling (dialnorm or the dialog level parameter), dynamic range compression (when invoked), and downmixing (summing multichannel audio appropriately for presentation on a stereo or mono playback system).

In all cases within licensed consumer decoders, and in most cases within professional licensed decoders (note that in some select professional receivers, special modes are available that allow for the disabling of the dialnorm subsystem; these special modes must be used with care), dialnorm processing is a mandatory process and is invoked on the decoded output at all times. When considering consumer and professional broadcast receivers, the decoded output of these units in most cases will be stereo signals, and when receiving multichannel programming, a cascade of processing occurs within the Dolby Digital (AC3) decoder. Specifically, 1) the audio is "leveled" according to the dialnorm parameter setting, 2) the dynamic range of the audio is controlled, and 3) the multichannel program is intelligently combined (downmixed) to create an appropriate stereo or mono signal.

It is important to understand that these processes work together. The dialnorm parameter identifies an area within the audio program that will not be processed by the dynamic range control subsystem. This "null band" provides a safe area for conversational speech. When the dynamic range control subsystem is invoked either through manual selection by the operator or, in the case of some consumer systems, by the manufacturer as a standard condition in order to protect small speakers from distortion and/or damage, audio elements that occur above the null band as identified by the dialnorm parameter are lowered in level and elements that occur below the null band are raised in level (Figure 2). The amount of dynamic range control is dictated by the mode selected by the operator and calculations made by the Dolby Digital (AC3) encoder that are embedded within the incoming data stream. Lastly, in the case of downmixing a multichannel signal to a stereo or mono presentation, the DRC subsystem, which in the case of downmixing is invoked automatically, helps to prevent clipping when these signals are combined.

If the dialnorm parameter is set such that it does not match the average level of conversational speech within the program, this cascade of processes can create audible artifacts as processing will take place on critical signals. Consider a program with the measured, average conversational speech level at -20dBFS but the dialnorm parameter is set to -31. In this case, the null band that is intended to protect critical conversational speech elements from inappropriate processing will be ineffective, and these critical speech elements will be pushed into areas of active dynamic range processing, putting the program at risk of audible compression artifacts and impacting viewer satisfaction.

A SYSTEMIC APPROACH

While the proper provisioning of broadcast reception and delivery equipment is a component of controlling loudness, the content itself must either contain the correct audio metadata or be provisioned such that the average level of speech within the content is either known or standardized to a consistent target level. One or both strategies can be implemented depending on the activity taking place and the equipment within the broadcast infrastructure.

Considering the varied delivery systems present in the typical broadcast infrastructure, it should be clear that program loudness cannot be successfully managed with a piecemeal approach. If locally produced content is properly managed but network signals, advertisements, and syndicated programming is not, level differences between programs will be objectionable, possibly resulting in viewer complaints. Additionally, when an analog feed is originated from a digital service, the incorrect selection of processing modes within the Dolby Digital (AC3) decoder may result in inappropriate audible artifacts in addition to level differences, also generating viewer complaints. Clearly what is necessary to achieve viewer satisfaction is a systemic approach which brings to bear equipment and workflow processes to achieve the desired result.

Live Programming

Controlling the audio program from acquisition to transmission provides the best opportunity to manage program loudness and eliminate unnecessary and inappropriate processing. For stereo or multichannel content that will be delivered using Dolby Digital (AC3), networks and local stations alike must decide upon a specific and consistent target level for conversational speech, and during the content creation process, use appropriate loudness metering during periods of conversational speech to ensure that the standardized average level of speech is maintained throughout the program. Often, the target level of speech is defined within a delivery specifications document. For stereo and multichannel programs, the target level of average conversational speech should match the specified level in the delivery specifications document. Again, when the dialnorm value within the Dolby Digital (AC3) data stream correctly matches the average level of speech, RF Mode DRC (aka narrow or heavy) will ensure that average speech levels will match existing NTSC programming, and the analog broadcast viewer will experience a seamless transition when channel surfing.

With respect to local stations, a consistent speech level is likely already being maintained, as there are fewer operators and the content is chiefly local news and special interest programs, both of which lend themselves to consistent procedures and practices. Empirical evidence suggests that production crews produce consistent results over time (Figure 3). As the data in Figure 3 suggests, any level differences in locally produced content for some stations may simply be the result of a shift change, and even that difference is likely to be slight. A medium-to-long term measurement strategy that includes proper metering will best diagnose any differences between production crews and provide the local station with a pathway to improve consistency.

For special events, the audio mixer must have a broadcast loudness meter in the audio booth, and throughout the event, must periodically monitor the measured result of the speech content, maintaining an artistic balance between program elements while also endeavoring to maintain a relatively consistent level with respect to speech. With the measured value of speech conveyed to the mixer via a clear, numeric display, consistent audio levels can be maintained while the mixer creates an appropriate balance between the audio elements, irrespective of operator fatigue.

Program Ingest

Content created for broadcast must meet specific standards, as described in the network or local delivery specifications. It is important that delivery specifications include an average level of speech along with a clear methodology to attain it so that a broadcaster can have some assurance that content received from outside providers for broadcast matches content created in house or from other outside vendors. Creating and publishing such a specification significantly reduces the chance of an externally produced program being rejected.

Here is an example of a loudness specification within a generic program delivery specification document:

The average level of dialog within the program on all non-compressed (i.e. non-Dolby E encoded) content (whether multichannel or stereo programs), evaluated using a Dolby LM100 Broadcast Loudness Meter or equivalent Leq(A) meter, shall return an average level of dialog of -27dBFS, +/-2dB measured over the entire length of the program. The LM100 should be set to "Infinite Mode" with "Dialog Intelligence" enabled. The measurement shall begin at the start of the program and stopped at the end of the program. The resulting Infinite Term measurement value shall be -27dBFS, +/-2dB.

Checking for adherence to published delivery specifications generally occurs at the point of ingest. When the delivery specifications contain a standardized average level of conversational speech and a clear methodology by which the content provider can attain it, these quality assurance (QA) processes must also include a check for the same specification using the same methodology. Similar to the Live Programming discussion earlier, the ingest operator should have an average loudness meter which is sensitive to speech in their toolkit, and understand how to use it. Additionally, the methodology outlined in the delivery specifications should be adhered to for the submitted content to stand the best chance of being approved for air. A detailed outline of measurement modes, settings and procedures will benefit both the content producer and the broadcaster.

Commercials and Ad Insertion

Television commercials have historically been blamed for most of the loudness problems within the medium. As Figure 4 suggests however, commercials in general, while loud, may appear to be quite consistent with respect to level. Considering the fact that commercial production techniques require that the "story" be told in a very short time period, this observed consistency with respect to level would seem to be a natural byproduct of the production process. Certainly they can be loud, in the case of Figure 4, averaging around -22dBFS, but the perceived loudness of any program will be judged in the home by the content preceding it. Herein lays the issue: If the program preceding the commercial break is perceived as louder than the commercials, the outlier will be the program content, while the opposite will be true if the program content is perceived as quieter than the commercials. The key is to either a) have each program segment (scripted programming, interstitials, news, commercials, etc.) contain a dialnorm value that matches the average speech level within it, or b) leverage the observed consistency of commercial content to match the level of the program content to the commercials.

Adjusting the level of commercial content may be a controversial task, as advertisers are paying for air time during a particular program. There are two strategies for tackling the issue of commercial loudness, and each may prove effective provided the appropriate content is measured prior to invoking a specific solution.

The first strategy involves either a) changing the level of the commercial content such that it matches a selected target level, or b) encoding the commercial content with the correct audio metadata. As commercials are often received locally via a fiber or satellite service and stored on servers prior to air, these programs can be provisioned automatically using non-real-time, file-based analysis and encoding products. These commercial files are handled by an automation or asset management system, and during their stay within a local station's IT network, can be analyzed and corrected through the use of faster-than-real-time, file-based processing engines, like the Dolby DP600 Program Optimizer. While this strategy processes commercial content and thus changes the level within, it is also automated and transparent, creating an attractive and cost effective solution to controlling program level.

These processing engines can also encode the file-based content prior to air and create the appropriate audio metadata that reflects the needs of the audio program within. In this fashion, the audio essence of the commercial is untouched thus removing the possibly controversial aspect of level adjustment, and the correct audio metadata is conveyed to the home where the Dolby Digital (AC3) decoder can properly provision it based upon the viewer's preference or the equipment's capabilities.

The second strategy is, in effect, the opposite of the previous one. In order to better match level differences between scripted programming and commercial content, the general consistency of commercial content observed in Figure 4 can be leveraged to provide an average standardized level for scripted programming. (Note: Each facility should measure and analyze a representative sample of its own commercial content rather than using the Figure 4 example as a basis for a level correction or processing strategy.) In this case, scripted programming would be required to match a level specified by the network or local station that matches the average level of commercial content. The burden of matching levels is placed on the content provider, but must be checked by the network or local station during ingest as described previously to ensure that it meets the mandated delivery requirements with respect to the average level of conversational speech. This strategy relies upon the assumption that the average level of every commercial will reside within a few dB of a selected target level. Because of the high probability of the level of some commercial content falling outside of this target acceptance window, there will likely be some instances of wider differentials between programs (i.e. average speech levels outside the target boundaries), and a higher incidence of viewer complaints, than if the first strategy is used exclusively.

Syndicated Programming

Like commercials, syndicated content is often received via a subscriber-based fiber or satellite delivery systems and stored on a server prior to air. Also like commercials, this delay in transmission can be used to the benefit of the broadcaster by allowing the content to be analyzed and corrected prior to air using non-real-time, file-based processing equipment.

Unlike commercials and ad content, syndicated programming does not benefit from a set of common production techniques. Syndicated programming can consist of chat shows, confrontational and reality programming, news and scripted programming. While chat shows and confrontational programming may use similar production techniques and audio processing equipment, scripted programming and even some reality-based programming generally involve a higher production value which in turn will result in a wider dynamic range.

Because of the wide variety of content available and the different production techniques used in syndicated content, it will not provide a generally consistent level to which other content can be matched. A local carrier can implement a manual leveling process involving a costly and burdensome series of checks and adjustments through an additional ingest path, but this is not a realistic solution for the modern broadcaster. In order to assure that syndicated content either a) carries an accurate dialnorm value that reflects the average level of conversational speech within the program, or b) the content itself matches a chosen target level for broadcast, the only realistic solution for managing program loudness on this type of programming is via non-real-time analysis, correction, and processing tools. These systems can perform leveling, encoding and trans-coding functions on various file types received by the local station, and do them transparently as an additional series of processes within the local station's normal IT network reception-to-transmission path.

Video on Demand Services

Premium programming can also benefit from non-real-time tools. When ingested onto the IT network, the program content can be analyzed and processed to provision it for air. In these instances, the content's audio essence can remain unchanged (i.e. not adjusted for level or limited in dynamic range) thus satisfying the viewers, program producers and any contractual obligations that may be in force. These non-real-time tools would be instructed to analyze the content and use the results of that analysis to create metadata values that will accurately reflect the content itself. Using these metadata values, the processes can then encode the content into Dolby Digital (AC3) and be ready for air, or create a broadcast wave file containing a metadata "chunk" which may be read by subsequent processing engines.

As with syndicated programming and commercial content, when a program resides within an IT network as a file, file-based analysis, correction and processing tools provide a "one stop shop" for managing loudness and provisioning content for air within a modern broadcast infrastructure.

CONCLUSION

Studies show that viewers will adjust the volume of their playback system in an effort to make the level of the speech element within a program consistent. Through the dialnorm parameter within audio metadata, Dolby Digital (AC3) provides a method to convey the measured average level of conversational speech within a program to the home, and when the dialnorm parameter is accurately set, dynamic range control subsystems within the Dolby Digital (AC3) decoder can appropriately provision the audio for the viewer's playback equipment or acoustic environment. When properly understood and utilized, the dynamic range subsystem within a Dolby Digital (AC3) decoder can be leveraged to provide an appropriate level match to existing analog content within a professional broadcast environment.

Loudness complaints from viewers are the result of perceived differences between program content. As such, even if 90% of a broadcaster's programming is level matched with respect to speech, viewer complaints will still occur due to the remaining 10% of unmatched content. For this reason, comprehensive loudness management and control requires a systemic approach that engages each program path within a broadcast infrastructure.

For live-to-air programming and content creation (news, sports, program interstitials, news teasers, etc.), identification of a target loudness level and placement of a broadcast loudness meter in the audio suite will provide real-time control as the content is being created. For program content that exists on a server (commercials, syndicated programming, video-on-demand programming, etc.), file-based, non-real-time tools provide the best method to both analyze and correct audio level issues as well as provision the audio content for air.

If networks, local stations, MSOs and content providers undertake a systemic approach to loudness control (Figure 5), viewer complaints can be reduced while simultaneously conveying wide dynamic range, compelling and engaging content which fulfills the promise of high definition television.

REFERENCES

1) EIA/CEA-CEB11, "NTSC/ATSC Loudness Matching," 2002.

2) Riedmiller, Lyman and, Robinson, "Intelligent Program Loudness Measurement and Control: What Satisfies Listeners?" presented at the 115th AES Convention, New York, 2003. https://aes2.org/publications/elibrary-page/?id=12415 

 

To access this document, please click here