FFmpeg: Extract Audio Only from Media File Using Best Presets

In response to your update: depending on how you're downloading the video, you may not be getting the best quality version that YouTube is hosting. If you're using clive (or cclive), -f best will get the best it can.

If you want to reliably not lose any data when pulling audio out of a video file, convert it to .wav or ensure you're not re-encoding.

You will first need to setup ALSA correctly by loading the snd_aloop module and using its named device in an mdev plugin. You will simultaneously output an application's audio through the loop back device and another device of your choice.

# ~/.asoundrc

pcm.!default {
  type plug
  slave.pcm mdev
  route_policy duplicate
}

pcm.mdev {
  type multi

  slaves.a.pcm "hw:Loopback,0,0"
  slaves.a.channels 2
  slaves.b.pcm "hw:0,0"
  slaves.b.channels 2

  bindings.0.slave a
  bindings.0.channel 0
  bindings.1.slave a
  bindings.1.channel 1
  bindings.2.slave b
  bindings.2.channel 0
  bindings.3.slave b
  bindings.3.channel 1
}

pcm.loopback {
  type hw
  card Loopback
  device 1
  subdevice 0
}

pcm.mic {
  type hw
  card 1
  device 0 
} 

FFMPEG can merge multiple audio streams using the amerge filter. You will provide it with the loop back source and you may provide additional inputs. This example includes a microphone. This script is setup for streaming to TwitchTV.

#!/bin/bash

FPS="30"
QUAL="faster"
GOP="60"
GOPMIN="30"
CBR="1500k"
ABR="96k"
SIZE="960x540"
OUTPUT=""rtmp://live-jfk.twitch.tv/app/${STREAM_KEY}"

function GetWindowPos() {
  echo $(xwininfo | grep 'Absolute' | awk '{ORS=" "; print $4;}' | awk '{print ":0.0+"$1","$2"+draw_mouse=0";}')
}

function GetWindowSize() {
  echo $(xwininfo | grep 'Width\|Height' | awk '{ORS=" "; print $2;}' | awk '{print $1"x"$2;}')
}

ffmpeg \
  -f alsa -i mic \
  -f alsa -i loopback \
  -f x11grab -s $(GetWindowSize) -i $(GetWindowPos) \
  -acodec aac -ab "$ABR" -strict -2 \
  -vcodec libx264 -preset "$QUAL" \
  -g "$GOP" -keyint_min "$GOPMIN" \
  -b:v "$CBR" -minrate "$CBR" -maxrate "$CBR" -bufsize "$CBR" \
  -filter_complex "[2:0]format=pix_fmts=yuv420p[vstream];[0:0][1:0]amerge=inputs=2[astream]" \
  -flags:v +global_header -flags:a +global_header \
  -r "$FPS" -s "$SIZE" \
  -map "[vstream]" -map "[astream]" \
  -f flv "$OUTPUT"

This would be an order of magnitude easier to use PulseAudio, but I often experience latency using the monitor sources, which is terrible for live streaming. If you are latency free on PulseAudio you could simply switch out the ALSA devices for two pulse inputs.

-f alsa -i mic \
-f alsa -i loopback \

These would change to:

-f alsa -i pulse \
-f alsa -i pulse \

Then you would set the inputs in pavucontrol.

• http://www.alsa-project.org/main/index.php/Matrix:Module-aloop
• https://ffmpeg.org/ffmpeg-filters.html