Version 0.1.0a2

Released December 21st, 2025

Added

Enabled in future versions after further testing, new audio streaming modes that offer a hybrid between RAM/storage audio streaming.
New undocumented voice gateway opcode 12. Awaiting further information from Discord.

AttributeError in FFmpegPool caused by an internal attribute not added to __slots__.

Released December 20th, 2025

Configuration override options in VoiceClient constructor. Allows the customization of output audio options, like bitrate and channels.
Volume settings per AudioPlayer, or override per source in each AudioSource constructor.
Clean VoiceClient.close() method.

Internal FFmpeg processes now pooled globally instead of per connection/guild. Detailed log below**.
Removed repetitive internal error catches - causing silent errors.

During testing/benchmarking, a singular FFmpeg spawned process per guild/connection would occupy ~4.5 MB of RAM on the host server.
Doing a bit of math, and a O(N) programmatic linear scale, 100 guild connections playing audio concurrently would occupy ~470 MB.
Because of that figure, and plans for near-future scalability, it was decided to make a FFmpeg process pool.
This pool would create new processes as they were needed, up to a certain amount per logical CPU core, up to a hardcap. All of this is configurable using the new configuration settings in the VoiceClient constructor.
Because these processes encode each audio source at once and then store the raw Opus frames in a buffer, each audio source goes from about 3.3 MB of MP3 to 2.6 MB of Opus inside of the buffer - with high quality settings - thus lowering the amount of RAM internally.
Taking an average of 10 processes at the same time, ~50 MB of RAM, and 100 servers streaming audio (2.6 MB of RAM each), it's about 310 MB of RAM used total, thus saving about 40% of RAM usage on a host server, dwindling hosting costs. 100 concurrent servers fits in the size of a Google Chrome Tab. Pretty neat.