I'm developing an application that has business logic that heavily depends on sound precision, hence I'm using the Web Audio API. The app has the functionality to manipulate audio files in multiple ways (fade in and fade out, programmatically seek through audios, play one audio multiple times, etc).
I've tried to implement this functionality using MediaElementAudioSourceNode, but I had a hard time getting everything together. Using this type of node, I wasn't sure how to implement some of the features like scheduling start time in the AudioContext timeline.
In the first iteration, I've implemented a simple download and decode method where the audio files are first downloaded as a whole from the network, stored to IndexedDB for caching purposes, then decoded into AudioBuffer and played as AudioBufferSourceNode connected to AudioContext.
As we already know, the decoding function is quite slow for larger files. After running some tests, I've realized that decoding OPUS is slower than decoding MP3. Also, decoding MP3 is slower than decoding WAV.
For that reason, in the second implementation, I've used decode-audio-data-fast and MP3 files. The decoding process is now faster as they're decoded in chunks on multiple threads, but I have to download bigger files.
However, I'd like to avoid downloading and decoding if possible. If this is not possible, I'm wondering if I can progressively download files and feed the chunks to the new WebCodecs API, then route them to AudioContext to be played and manipulated. The files are hosted on an S3 bucket. In theory, as far as I know, this should be completely possible, however, due to a lack of examples and documentation, and probably my experience with this, I can't figure out how exactly to implement this.