HLS super fast frame accuracy trimming (JS)

#Usage

• Install: ffmpeg and nodejs
• Ensure ffmpeg is in the PATH
• Use this syntax: ./trim_hls_frame_acc source_hls dest_file tmp_dir trim_in trim_out
  • Example1: ./trim_hls_frame_acc /hls_test/ /out/test.mp4 /tmp 10.0 21.2
  • Example2: ./trim_hls_frame_acc /hls_test/test.m3u8 /out/test.mp4 /tmp 10.0 21.2

#Description

• This code is an evolution of this code: [Trim HLS stream with frame accuracy using ffmpeg and Bash script](https://jordicenzano.name/2014/08/07/trim-hls-stream-with-frame-accuracy-using-ffmpeg-and-bash-script-2/)
• The purpose of this code gets a group if HLS chunks (.ts) and input and output timestamps and creates an output mp4 file frame accuracy trimmed, AV perfectly aligned, and thanks to the underlying algorithm in a very small amount of time.
• It can be useful to:
  • On the fly live streaming trimming (highlights)
  • VOD frame accuracy trimming
  • Increase dramatically the speed of video editors (mostly cloud video editors)

#How it works

1. Based on the in an out points analyze the input files an establish which are the first and last segments
2. Split A/V of the necessary chunks, creating a .ts (video only) and .acc files. It is just a parsing operation.
3. For the first and the last chunks:
   • Get video frame position information from and calculate the video cutting point to preserve frame accuracy
   • Decode to YUV
   • Encode the trimmed YUV to the same codec as the original files
4. Concatenate the video segments to a final video .ts
5. For the first and the last audio chunks:
   • Get audio frame position information and calculate the audio cutting points based on video cutting points to preserve A/V alignment
   • Calculate AV delay necessary at the output file to preserve AV alignment
   • Trim the stream at packet level
6. Concatenate the audio segments to a final audio .aac
7. Mux final video and final audio file to MP4 preserving the previously calculated A/V delay

#Assumptions

• The video codec is H264 and the audio coded is aac
• All the chunks start with and I frame (wowza forces an I frame at start of chunks, event the transmuxed ones)
• Only 1 video and 1 audio present in the input files
• If you use hls manifest (.m3u8) as a source, only relative local paths are allowed
• The A/V delay in the original stream is reasonable (<2s)

#Future work

• Make the code usable for any input different than HLS
• Apply multi-threaded processing in some stages to speed up the process (reasonable goal is to increase the speed x2)
• Modify the code to process the input files at video GOP level (not chunk level)
  • Increase the processing speed (GOPs use to be smaller than chunks)
  • No need that chunks start with I frame
  • Easier to handle other formats different than HLS