ManagedCodecs.Aac 0.2.1

ManagedCodecs

Pure managed C# media decoders: no native binaries, Native AOT friendly, and running on WebAssembly too. Every decoder is checked against FFmpeg sample by sample.

• H.264: Constrained Baseline to High 4:4:4 Predictive. CAVLC and CABAC, I/P/B, interlaced (MBAFF and field pictures), 4:0:0 to 4:4:4, 8 to 14 bits, lossless; bit-exact with FFmpeg. Frame threading and error concealment on request.
• MJPEG: baseline and extended Huffman JPEG, interlaced AVI1; bit-exact with FFmpeg.
• AAC LC and HE-AAC (SBR, parametric stereo): mono to 7.1, within 2^-17 of FFmpeg's float output for AAC LC.
• Containers: AVI, MP4 (fragmented files and edit lists too) and raw AAC (ADTS), packets and timestamps as FFmpeg's demuxers give them.

The packages: ManagedCodecs for everything, or one of ManagedCodecs.Containers, ManagedCodecs.Mjpeg, ManagedCodecs.H264 and ManagedCodecs.Aac, each with ManagedCodecs.Core. They share internals, so each depends on the others at exactly its own version.

Decoding a file

Packets go in in decode order; frames come out in display order, each with the Pts of its packet. Decode returns zero or more frames (H.264 can hold frames back for reordering) and Flush the rest:

using ManagedCodecs;
using ManagedCodecs.Containers;
using ManagedCodecs.H264;
using ManagedCodecs.Mjpeg;

using var demuxer = Demuxer.Open("video.mp4");
var video = demuxer.Video!;
IVideoDecoder decoder = video.Codec == CodecIds.H264 ? new H264Decoder(video) : new MjpegDecoder(video);

foreach (var packet in demuxer.ReadPackets().Where(p => p.StreamIndex == video.StreamIndex))
{
    foreach (var frame in decoder.Decode(packet))
    {
        Show(frame); // frame.Pts is in video.TimeBase; frame.GetRow(plane, y) reads its samples
    }
}

foreach (var frame in decoder.Flush())
{
    Show(frame);
}

Publishing with Native AOT on x64: set <IlcInstructionSet>x86-64-v3</IlcInstructionSet> (AVX2) in the app, or the 256-bit paths are left out and AAC and HE-AAC take about a third longer (README, Native AOT).

Audio goes the same way: new AacDecoder(demuxer.Audio!) returns an AudioFrame of float samples per packet. Every decoder and demuxer also has an async API (ReadPacketsAsync, DecodeAsync, FlushAsync).

More

The repository has the full documentation: the README (the API in detail, threading, malformed input, WebAssembly, speed against FFmpeg), what each feature is checked against, and the changelog. MIT licensed.