Why xHE-AAC is being embraced at Meta

• We’re sharing how Meta delivers high-quality audio at scale with the xHE-AAC audio codec.
• xHE-AAC has already been deployed on Fb and Instagram to supply enhanced audio for options like Reels and Tales. 

At Meta, we serve each media use case possible for billions of individuals the world over — from short-form, user-generated content material, equivalent to Reels, to premium video on demand (VOD) and reside broadcasts. Given this, we’d like a next-generation audio codec that helps a variety of working factors with wonderful compression effectivity and trendy, system-level audio options. 

To handle these wants now and into the longer term, Meta has embraced xHE-AAC because the car for delivering high-quality audio at scale.

The advantages of xHE-AAC

xHE-AAC is the newest member of the MPEG AAC audio codec household. The Fraunhofer Institute for Integrated Circuits IIS performed a considerable function within the growth of xHE-AAC and the MPEG-D DRC customary.

At present, xHE-AAC is already offering a superior audio expertise on Fb and Instagram — together with on Reels and Tales — and has a lot of beneficial options.

Loudness administration

With tons of of hundreds of thousands of uploads per day throughout Fb and Instagram, we obtain audio tracks with loudness ranges starting from silence to full scale, and all the pieces in between. 

When individuals play these movies sequentially, they will understand some audio as being too loud or too quiet. This creates listener fatigue from having to consistently alter the quantity.

xHE-AAC’s built-in loudness administration system solves for loudness inconsistency whereas meticulously preserving creator intent by bringing the common loudness of all classes to the identical goal degree and managing the dynamic vary of every session to suit the playback setting.

As a substitute of burning in a selected goal degree and dynamic vary compression (DRC) profile throughout encoding, xHE-AAC permits us to depart the unique audio traits untouched and delegate loudness administration processing to the consumer through loudness metadata, for the optimum audio expertise based mostly on context. 

Because of xHE-AAC’s loudness administration, individuals can spend extra time immersed of their favourite content material and fewer time twiddling with the quantity management.

Adaptive bit fee audio

Most individuals who use our apps eat media on cell gadgets and count on the very best audio high quality with out interruption. This presents a problem for streaming media as a result of connection high quality varies on cell and can lead to a really uneven person expertise. 

To optimize high quality below dynamic bandwidth constraints, we produce a number of video and audio qualities to match various community situations at playback time. Though we produce a number of audio lanes, now we have traditionally solely employed adaptive bit fee (ABR) algorithms to change video qualities throughout playback as a result of it’s troublesome to allow adaptive bit fee audio with out compromising high quality throughout lane transitions.

To be able to allow seamless audio ABR, xHE-AAC introduces the idea of quick playout frames (IPFs) that comprise all the information vital to begin taking part in a brand new audio lane with out counting on information from different frames. By putting an IPF at the start of every Dynamic Adaptive Streaming over HTTP (DASH) phase and aligning the phase durations of every lane, we will seamlessly swap between audio lanes throughout playback to supply the highest-quality audio at any accessible bandwidth whereas avoiding playback stalls.

After launching audio ABR on Fb for Android, we have been in a position to enhance person expertise by lowering the variety of classes the place playback stalls. 

How we deployed xHE-AAC

We generate xHE-AAC bitstreams utilizing an encoder SDK supplied by the Fraunhofer Institute for Built-in Circuits IIS, after which put together the ensuing audio recordsdata for DASH streaming with shaka-packager. The xHE-AAC encoder’s two-pass encoding mode is used to measure the enter loudness envelope and common program loudness on the primary cross and carry out the precise audio information compression on the second cross. As an additional advantage, two-pass encoding permits us to make use of loudness vary management (LRAC) DRC, which mitigates pumping artifacts in any other case launched by single-pass DRC algorithms.

To arrange an xHE-AAC audio adaptation set for ABR supply, IPFs are inserted at fixed time intervals, audio configuration parameters equivalent to pattern fee and channel configuration are stored fixed, and distinctive stream identifiers are chosen for every lane within the audio adaptation set.

At playback time, we custom-fit the audio to the listening setting by configuring a goal loudness degree and DRC impact sort based mostly on context, and due to the embedded loudness metadata, we will adapt a single xHE-AAC bitstream to a wide range of audio consumption use circumstances, from headphones to system audio system and varied ranges of background noise. Lastly, if the consumer is starved for information or bandwidth is plentiful, audio ABR will routinely swap audio qualities to make sure that the very best audio high quality is performed with out interrupting the playback session.

The place are you able to expertise xHE-AAC at present?

You’ll be able to expertise xHE-AAC audio on Fb for iOS and Android, in addition to on focused surfaces on Instagram, equivalent to Reels and Tales. We encourage you to put in the newest model of Fb and Instagram apps on iOS 13+ and Android 9+ to make sure which you could expertise it.

Acknowledgements

This work is the collective results of your complete Video Infrastructure and Instagram Media Platform groups at Meta in collaboration with Fraunhofer Institute for Built-in Circuits IIS. The writer wish to prolong particular because of Abhishek Gera, Tim Harris, Arun Kotiedath, Edward Li, Meng Li, Srinivas Lingutla, Denise Noyes, Mohanish Penta, David Ronca, Haixia Shi, Mike Starr, Cosmin Stejerean, Jithin Parayil Thomas, Simha Venkataramaiah, Juehui Zhang, Runshen Zhu, and the engineering group at Fraunhofer Institute for Built-in Circuits IIS.