Abstraction for streaming raw PCM audio in browser via Web Audio API.
npm install audio-feederaudio-feeder
============
The AudioFeeder class abstracts a buffered output pipe for uncompressed PCM
audio in the browser, supporting the standard W3C Web Audio API.
AudioFeeder was written for the ogv.js in-browser Ogg/WebM media player,
and is suitable for use in custom audio and video playback.
* main AudioFeeder & Web Audio code path under MIT license, (c) 2013-2021 Brion Vibber
options.backendFactory.initSharedAudioContextaudioNode option allows attaching to non-default destinationstop()/start() buffered audio recoverystop()/start() more reliable, doesn't drop audioplaybackPosition no longer advances while pausedstop()/start()flush() method; use to clear buffers after a stop when seeking etconstarved callback for Flash backendonbufferlow callback when buffered data gets low, but not yet emptybufferThreshold property to get/set the threshold in secondsdurationBuffered property to track amount of data left to playplaybackPosition property mirroring getPlaybackState().playbackPositionIf your project is built with webpack or browserify, it's easy to bundle up
AudioFeeder's JavaScript classes; you will have to manually ensure that the
Flash shim for IE is bundled alongside it.
Add to your npm dependencies:
``
npm install audio-feeder
and in your using code, set up the class like so:
`
var AudioFeeder = require('audio-feeder');
Grab AudioFeeder.js or AudioFeeder.min.js (minified) from the ZIP download or
from dist/ subdir in the npm module.
Include either as a module (CommonJS or AMD) or a standalone script.
`js
// Create a feeder object
var feeder = new AudioFeeder();
// Set up 2-channel stereo, 48 kHz sampling rate
feeder.init(2, 48000);
// Flash mode for IE 10/11 requires waiting.
feeder.waitUntilReady(function() {
// Buffer some data before we start playback...
//
// Each channel gets its own 32-bit float array of samples;
// this will be 0.25 seconds of silence at 2ch/48kHz.
//
// Note it's ok for each bufferData() call to have a different
// number of samples, such as when working with a data format
// with variable packet size (Vorbis).
//
feeder.bufferData([
new Float32Array(12000),
new Float32Array(12000)
]);
// Start playback...
feeder.start();
document.querySelector('button.stop').addEventListener('click', function() {
// You can pause output at any time:
feeder.stop();
// to release resources, call feeder.close() instead.
});
// Callback when buffered data runs below feeder.bufferThreshold seconds:
feeder.onbufferlow = function() {
while (feeder.durationBuffered < feeder.bufferThreshold) {
feeder.bufferData([
new Float32Array(12000),
new Float32Array(12000)
]);
}
};
});
`
See also the included demo.html file for a live sample web page.
WebAudio-specific:
* audioContext: an AudioContext object to be use instead of creating a new oneoutput
* : an AudioNode object to attach output to instead of the default
General:
* backendFactory: a function to call to provide a custom backend, for testing or advanced usage. Should take params numChannels, sampleRate, and options and return an object conforming to the internal backend class protocol. This is not considered a stable API at this time.
AudioFeeder works with 32-bit floating point PCM audio. Data packets are
represented as an array containing a separate Float32Array for each channel.
Warning: this may change to use a wrapper class before 1.0.
The current playback position in seconds, and the duration of buffered but not
yet played data, are available through the playbackPosition anddurationBuffered properties.
Additional playback state can be retrieved from the getPlaybackState() method:
`js`
{
playbackPosition: Float / seconds of sample data that have played back so far /,
samplesQueued: Float / samples remaining before the buffer empties out, approximate /,
dropped: Integer / count of buffer underrun events /,
delayed: Float / total seconds of silence played to cover underruns /
}
Warning: this structure may change before 1.0.
playbackPosition tracks the time via actual samples output, corrected for drops
and underruns. This value is suitable for use in scheduling output of synchronized
video frames.
This high-level pseudocode shows a simplified version of the playback sync logic
from the ogv.js video player:
`js
function processMediaData() {
while (codec.audioReady && audioFeeder.durationBuffered < audioFeeder.bufferThreshold) {
// When our audio buffer gets low, feed it some more audio data.
audioFeeder.bufferData(decodeAudioPacket());
}
if (codec.frameReady && audioFeeder.playbackPosition >= codec.nextFrameTimestamp) {
// When the audio playback has reached the scheduled time position
// of the next frame, decode and draw it.
player.drawFrame(codec.decodeVideoPacket());
}
// And check back in before the next frame!
if (codec.dataPending) {
requestAnimationFrame(processMediaData);
}
}
// Fire off an animation-based loop...
requestAnimationFrame(processMediaData);
// If in a background thread, animation loops will be throttled.
// Also fire when audio gets low!
audioFeeder.onbufferlow = processMediaData;
`
The caller is responsible for maintaining a loop and scheduling any decoding,
frame drawing, etc.
Beware that setTimeout, setInterval, and requestAnimationFrame may be throttled
on background tabs, leading to spotty performance if scheduling decoding
based on them.
To avoid background tab throttling, use the onbufferlow event callbackbufferThreshold
to run additional decoding/buffering. This is fired asynchronously when
the available buffered data runs below seconds.
You can buffer an arbitrarily large amount of audio data, but for non-trivial
examples it's best to decode or generate audio in smallish chunks and buffer
them over time. Pre-buffering will eat more memory, and could lead to slowness
on the main thread if you process a lot of data on the main thread in one
function call.
Performing other slow tasks on the foreground thread may also prevent the
Web Audio API callbacks from being called in a timely fashion, resulting in
audio underruns even if lots of data has been buffered up.
There are currently two supported events, set via the 'onstarved' and
'onbufferlow' properties.
'onstarved' is called when buffered data runs out during playback,
giving a last-chance opportunity to buffer more data. This is a synchronous
call in the audio path, and may not be enough to guarantee good performance.
'onbufferlow' is called asynchronously when the buffered data runs
lower than a configurable threshold, which is more flexible. This
threshold is available for get and set via the bufferThreshold property,
defaulting to twice the low-level buffer duration.
You can use these events to buffer additional data at the last minute,
or to trigger a close-out of the feeder when no more data is available.
Todo:
* add events for beginning of playback?
* add event for scheduled end of playback
An earlier version of AudioFeeder supported a Flash backend for supporting
Internet Explorer 10/11. This is no longer possible since the Flash plugin
was disabled by Adobe in 2021.
As of 0.4.21, the tempo or playback rate can be modified without altering pitch, suitable for use in implementing video playback control.
Set the tempo property to a value larger than 1 to multiply speed, or less than 1 to decrease it. The playbackPosition property is returned in input units, making it fairly easy to report back a current time suitable for A/V sync; however some properties such as durationBuffered will return output units and should be used carefully when at non-default empos.
Currently the tempo control is not applied to data that has already been buffered, which can produce a "lag time" before tempo changes take effect.
Note that stereo or multi-channel input is mixed down to mono for processing when the tempo is not set at 1; see audio-tempo-changer#1.
The pre-packed AudioFeeder.js included in ZIP and npm releases can be rebuilt
from the source files. This is known to work on Mac, Linux, and Windows.
Build prerequisites:
* node.js / npm
`bashFetch build dependencies (webpack, eslint etc)
npm install
This will produce a 'dist' subdirectory containing a ready to use
AudioFeeder.js, and AudioFeeder.min.js as well as a demo.html example page.