From 515a9f479a717c80ad7fe800d6f9d13f32c7e601 Mon Sep 17 00:00:00 2001
From: Michael Campagnaro <mikecampo@protonmail.com>
Date: Thu, 22 Jun 2023 15:55:57 -0400
Subject: [PATCH] Improve video audio normalization scripts

---
 dotfiles/bin/extract-16bit-wav-from-video |  1 +
 dotfiles/bin/fix-audio-in-one-channel     |  2 +-
 dotfiles/bin/normalize-video-volume       | 33 ++++++++++++++---------
 dotfiles/bin/remove-audio-from-video      |  2 +-
 dotfiles/bin/trim-accurate-video          |  2 ++
 dotfiles/bin/trim-video                   |  2 ++
 6 files changed, 27 insertions(+), 15 deletions(-)

diff --git a/dotfiles/bin/extract-16bit-wav-from-video b/dotfiles/bin/extract-16bit-wav-from-video
index 32802c9..b92d68f 100644
--- a/dotfiles/bin/extract-16bit-wav-from-video
+++ b/dotfiles/bin/extract-16bit-wav-from-video
@@ -45,6 +45,7 @@ fi
 
 printf "\n${YELLOW}${BOLD}Extracting 16-bit WAV from $input | output: $output_name${NORMAL}\n"
 
+# -ac 1 mixes audio to a single channel.
 ffmpeg -i "$input" -ar 16000 -ac 1 -c:a pcm_s16le "$output_name"
 
 printf "${GREEN}${BOLD}Done extracting 16-bit WAV from $input | output: $output_name${NORMAL}\n"
diff --git a/dotfiles/bin/fix-audio-in-one-channel b/dotfiles/bin/fix-audio-in-one-channel
index 17c5e00..c795523 100644
--- a/dotfiles/bin/fix-audio-in-one-channel
+++ b/dotfiles/bin/fix-audio-in-one-channel
@@ -45,7 +45,7 @@ fi
 
 printf "\n${YELLOW}${BOLD}Repairing audio in $filename.$extension | output: $output${NORMAL}\n"
 
-ffmpeg -i "$filename.$extension" -c:v copy -ac 1 "$output"
+ffmpeg -i "$filename.$extension" -c:v copy -ac 1 -map 0 "$output"
 
 printf "\n${GREEN}${BOLD}Done repairing audio in $filename.$extension | output: $output${NORMAL}\n\n"
 
diff --git a/dotfiles/bin/normalize-video-volume b/dotfiles/bin/normalize-video-volume
index 2580a25..0628636 100644
--- a/dotfiles/bin/normalize-video-volume
+++ b/dotfiles/bin/normalize-video-volume
@@ -1,6 +1,8 @@
 #!/usr/bin/env bash
 
-# Use this to normalize the audio of a video using the average loudness, or RMS-based normalization.
+# Use this to normalize the audio of a video using the average loudness, or RMS-based normalization. It does a pretty good job!
+# If you want to modify the volume yourself then checkout the analyze-video-volume and change-video-volume scripts.
+#
 # This does not re-encode the video.
 #
 # Inspired by https://superuser.com/a/323127 and https://superuser.com/a/1312885
@@ -50,21 +52,26 @@ printf "\n${YELLOW}${BOLD}Normalizing audio in $filename.$extension | output: $o
 # This is done in two passes. The first pass will compute the mean loudness and
 # the second pass will normalize the audio using the mean as the target.
 
-# -vn, -sn,  and -dn tells ffmpeg to ignore non-audio streams during the analysis. This speeds things up.
-ffmpeg -i "$filename.$extension" -af "volumedetect" -vn -sn -dn -f null /dev/null
+temp_file="/tmp/_audio_info_$RANDOM"
 
-#ffmpeg -i "$filename.$extension" -c:v copy -ac 1 "$output"
+# 1st pass:
+cmd="ffmpeg -i \"$filename.$extension\" -pass 1 -filter:a loudnorm=print_format=json -vn -sn -dn -f null /dev/null 2>&1 | sed -n '/{/,/}/p' > $temp_file"
+printf "\n${BOLD}Running 1st pass:\n$cmd\n\n${NORMAL}"
+eval $cmd
 
-printf "\n${GREEN}${BOLD}Done normalizing audio in $filename.$extension | output: $output${NORMAL}\n\n"
+printf "\n${BOLD}${GREEN}Done.\n\n${NORMAL}"
 
+ii=`grep   \"input_i\"       $temp_file | cut -d: -f2 | tr -cd [:digit:].-`
+itp=`grep  \"input_tp\"      $temp_file | cut -d: -f2 | tr -cd [:digit:].-`
+ilra=`grep \"input_lra\"     $temp_file | cut -d: -f2 | tr -cd [:digit:].-`
+it=`grep   \"input_thresh\"  $temp_file | cut -d: -f2 | tr -cd [:digit:].-`
+to=`grep   \"target_offset\" $temp_file | cut -d: -f2 | tr -cd [:digit:].-`
 
+# 2nd pass:
+cmd="ffmpeg -i \"$filename.$extension\" -c:v copy -pass 2 -filter:a loudnorm=linear=true:I=-16:LRA=11:tp=-1.5:measured_I=$ii:measured_LRA=$ilra:measured_tp=$itp:measured_thresh=$it:offset=$to:print_format=summary -map 0 \"$output\""
+printf "${BOLD}Re-encoding audio:\n$cmd\n\n${NORMAL}"
+eval $cmd
 
-#---------------------------------------
-# This seems better. 2 pass using loudnorm filter.
+printf "\n${GREEN}${BOLD}Done normalizing volume in $filename.$extension | output: $output${NORMAL}\n"
+rm $temp_file
 
-# 1st pass: ffmpeg -i "$filename.$extension" -pass 1 -filter:a loudnorm=print_format=json -vn -sn -dn -f null /dev/null
-# 2nd pass: ffmpeg -i "$filename.$extension" -c:v copy -pass 2 -filter:a loudnorm=linear=true:measured_I=$input_i:measured_LRA=$input_lra:measured_tp=$input_tp:measured_thresh=$input_thresh "$output"
-
-# TODO: extract the $input_i, input_lra, etc from the 1st pass output so that this can be automated.
-# TODO: stackoverflow said if there are subtitles or multiple vid streams then add "-map 0" before the output name. Test this.
-# TODO: disable the log file or just delete it after normalizing.
diff --git a/dotfiles/bin/remove-audio-from-video b/dotfiles/bin/remove-audio-from-video
index a331e68..94d3c57 100644
--- a/dotfiles/bin/remove-audio-from-video
+++ b/dotfiles/bin/remove-audio-from-video
@@ -43,7 +43,7 @@ fi
 printf "\n${YELLOW}${BOLD}Removing audio from '$filename.$extension' | output: '$output'${NORMAL}\n"
 
 # -an removes the audio.
-ffmpeg -i "$filename.$extension" -c:v copy -an "$output"
+ffmpeg -i "$filename.$extension" -c:v copy -an -map 0 "$output"
 
 printf "\n${GREEN}${BOLD}Done removing audio from '$filename.$extension' | output: '$output'${NORMAL}\n\n"
 
diff --git a/dotfiles/bin/trim-accurate-video b/dotfiles/bin/trim-accurate-video
index 6bd29c6..a2538cf 100644
--- a/dotfiles/bin/trim-accurate-video
+++ b/dotfiles/bin/trim-accurate-video
@@ -42,6 +42,8 @@ timing_args="-ss $start_time -to $end_time"
 
 printf "\n${YELLOW}${BOLD}Trimming '$filename.$extension' | output: $output | start: $start_time | end: $end_time${NORMAL}\n"
 
+# You might have issues if the file has multiple video streams or embedded subtitles. The -map 0 arg is typically given
+# when copying a video stream, but I'm not sure if it's appropriate to use here.
 ffmpeg -y -stats -loglevel level+error $timing_args -accurate_seek -i "$filename.$extension" -c:v libx264 -c:a copy "$output"
 
 printf "\n${GREEN}${BOLD}Finished trimming${NORMAL}\n\n"
diff --git a/dotfiles/bin/trim-video b/dotfiles/bin/trim-video
index 7d54049..7388b11 100644
--- a/dotfiles/bin/trim-video
+++ b/dotfiles/bin/trim-video
@@ -40,6 +40,8 @@ timing_args="-ss $start_time -to $end_time"
 
 printf "\n${YELLOW}${BOLD}Trimming '$filename.$extension' | output: $output | start: $start_time | end: $end_time${NORMAL}\n"
 
+# You might have issues if the file has multiple video streams or embedded subtitles. The -map 0 arg is typically given
+# when copying a video stream, but I'm not sure if it's appropriate to use here.
 ffmpeg -y -stats -loglevel level+error $timing_args -i "$filename.$extension" -c copy "$output"
 
 printf "\n${GREEN}${BOLD}Finished trimming${NORMAL}\n\n"