Live Dynamic Loudness Demo

• Algorithm based on ITU-R BS.1770
  • At time of writing, version 5 (11/2023) was the latest
• Implemented in plain javascript using Web Audio API

Usage

• Clone repository and open index.html file in Browser (tested with FF 60.1.0esr) or start webserver
• Audio files sample including test files from ITU-R-BS.2217 and a couple of others located under /files subdirectory
  • • At time of wrinting, version 2 (10/2016) was the latest
• Open audio file via Select File Button
  • Depending on the files size and format this may take a little
• Adjust parameters if needed and Play
• You can also Reset Loudness Calculation. Can e.g. simulate some artificial track change while playing
• Or visit github pages: https://goepfert.github.io/loudness_adaption_demo/
  • But you would probably need some audio files (e.g. from this repo)

Remarks

Certainly needs some more testing if properly implemented, but doesn't look so bad in the first place. It survived all tests for up to two channels from ITU-R-BS.2217. Hint: Crank up Window Size to be large than the file size to check for file loudness at the end of a file. And also some refactoring and code review ...

Parameters

Interval Size The file (stream) will be segmented in time intervals of given size. This is the smallest segment a loudness can be calculated. Defaults to 400 ms.

Overlap between Intervals The Intervals can overlap each other. Defaults to 75%.

Window Size How many seconds the algorithm shall look back if enough data is available. Defaults to 8 seconds

Signal Flow

AudioBufferSource -> Live Loudness Calculation (Web Audio Worklet)
                               -> Gain Control -> Post Loudness Calculation (Web Audio Worklet)
                                                          -> Volume Meter (Web Audio Worklet)
                                                          -> Audio Out

What is plotted

Live Loudness The current calculated loudness from the start of the file (if not in loop) or since last reset (Button: Reset Loudness Calculation). The total time is limited to the Window Size parameter.

Target Loudness The target loudness in LKFS before any audio processing.

Target Gain The gain to compensate the difference between Live Loudness and Target Loudness. The target gain is directly applied with an exponential ramp and a decay constants that can be parametrised in dB/s. (The timeConstant is set to get 95%, for details see here). The gain correction can be bypassed by unchecking the checkbox apply gain orrection. Remark: There is no clipping control implemented.

Loudness after gain control The Loudness is (re)calculted after setting the target gain with a fixed Window Size that can be different as for Live Loudness calculation.

Volume Meter

A simple attempt to check if there is clipping. It checks if there are samples with values near the maximum (abs(value) > 0.98) within the last 250 ms. If so, it meter graph turns red.

TODOs

• Implement and test multi-channel audio (besides stereo).
• Check further files and prepare file with simulated source change.
• Implement possibility to do offline file loudness calculation, i.e. calculate file loudness without actually playing the file.
• Go away from deprecated ScriptProcessorNode
  • Done, refactored to Audio Worklet including a couple of performance optimizations (Master branch)
  • Old ScriptProcessorNode still available in branch Scriptprocessor_deprecated