Strategies for Per-Sample Processing of Audio Graphs in the Browser

Abstract

Due to current browser limitations, most synthesis in the browser is currently performed using the block-rate nodes included in the WebAudio API. However, block-rate processing of audio graphs precludes many types of synthesis in addition to limiting both the accuracy and flexibility of scheduling. We describe alternative strategies for performing effi cient, per-sample processing of audio graphs in the browser using the ScriptProcessor node, affording synthesis techniques that are not commonly found in existing Java- Script audio libraries. We introduce a new library, Genish.js, that provides unit generators for common low-level synthesis tasks and acts as a compiler for signal processing functions; this library is a loose port of the Gen framework for Max/MSP. We used Genish.js to update a higher-level library for audio programming, Gibberish.js, realizing im- provements to both e ciency and audio quality. Preliminary benchmarks comparing the performance of Genish.js audio graphs to equivalent graphs made with the WebAudio API show promising results.