MediaElementSourceNode should have start/stop methods ala AudioBufferSourceNode

[luser]

The spec says that AudioBufferSourceNode "is useful for playing short audio assets which require a high degree of scheduling flexibility". For longer assets it seems to encourage using MediaElementSourceNode. However, you then lose the ability to precisely schedule playback, which seems unfortunate. I started looking into writing a multitrack recorder using the Web Audio API, but this seems like a bit of a showstopper since it doesn't seem possible to precisely schedule the playing of longer samples.

[notthetup]

I have used AudioBufferSourceNode for > 10min long assets and it works totally fine. You're only restricted by the size of memory your browser can allocate.

[luser]

Be that as it may, it seems silly to restrict precise scheduling to assets that fit in memory. Having poked at Gecko's media internals in the past I suspect this would not be a hard thing to support.

[padenot]

It's straightforward to precisely schedule an asset in the future, as you mention.

AudioBufferSourceNode's goal it to be able to start playing a buffer as fast as possible in the future, (say, possibly in reaction to user interaction). Any kind of network or disk IO is prohibited, in this context.

[joeberkovitz]

Resolution: Clarify spec, ABSN is useful for more than short assets. If sample accurate scheduled streaming required, currently should roll your own with AudioWorker.

[billhofmann]

So, language seems like it should note that, rather than "short assets", it's memory only assets, and assets that require network or disk I/O should be scheduled using AudioWorker.

[billhofmann]

Clarify use of AudioBufferSourceNode and provide recommendation for non-memory backed assets

[dorontal]

I cannot see how AudioWorker (I assume you mean an Audio Worklet Node) could help the start playback precise scheduling issue. By the time an audio worklet node gets its audio, that audio has already been scheduled - it has already started playing. But we need to control the timing of precisely when the playback starts. How is an Audio Worker going to help?

[hoch]

MESN and other Media-related AudioNodes are backed by non-memory assets. Thus scheduling them like AudioBufferSourceNode would be difficult. For that, we need a new scheduling model for real-time streaming backend and that is outside of V1 work.

This might be something in the area of V2 decoding/encoding work. FWIW, we triaged it as "High Priority V2" issue.