Merge branch 'master' of github.com:kanjieater/AudiobookTextSync

xingbicheng · Mar 23, 2023 · 22449e5 · 22449e5
2 parents 304e212 + aa22a38
commit 22449e5
Showing 1 changed file with 5 additions and 4 deletions.
diff --git a/readme.md b/readme.md
@@ -36,11 +36,12 @@ Primarily I'm using this for syncing audiobooks to their book script. So while y
 2. Make sure you run any commands that start with `./` from the project root, eg after you clone you can run `cd ./AudiobookTextSync`
 1. Setup the folder. Create a folder to hold a single media file (like an audiobook). Name it whatever you name your media file, eg `Arslan Senki 7`, this is what should go anywhere you see me write `<name>`
 2. Get the book script as text from a digital copy. Put the script at: `./<name>/script.txt`. Everything in this file will show up in your subtitles. So it's important you trim out excess (table of contents, character bios that aren't in the audiobook etc)
-3. You need _both_ the audiobook as a full m4b (technically other formats would work), AND the split parts. As long as you have one, you can easily get the other. You could technically only use the full single file, but you will most-likely run out of ram for longer works. See [Whisper ~13GB Memory Usage Issue for 19hr Audiobook](https://github.com/jianfch/stable-ts/issues/79). By using small splits, we can have more confidence the Speech To Text analysis won't get killed by an Out Of Memory error.
-4. Split files should be `./<name>/<name>_splitted/`.If you have the full audiobook as a m4b, you can split it into chapters using `./split.sh "<full folder path>"`. eg `./split.sh "/mnt/d/Editing/Audiobooks/かがみの孤城/"`
+3. If you have less than ~8GB of free RAM you may need _both_ the audiobook as a full m4b (technically other formats would work), AND the split parts. As long as you have one, you can easily get the other. You could technically only use the full single file, but you may run out of ram for longer works. See [Whisper ~13GB Memory Usage Issue for 19hr Audiobook](https://github.com/jianfch/stable-ts/issues/79). By using small splits, we can have more confidence the Speech To Text analysis won't get killed by an Out Of Memory error. As of this tool using stable-ts 2.0.0, this has been improved significantly. I no longer recommend using split files, as it's more likely to introduce a ~50ms delay in the subtitles. It's shouldn't be an issue if you use MPV and can correct the subtitle timings with a hotkey press though.
+4. If you are not using split files you can skip this step. Split files should be `./<name>/<name>_splitted/`.If you have the full audiobook as a m4b, you can split it into chapters using `./split.sh "<full folder path>"`. eg `./split.sh "/mnt/d/Editing/Audiobooks/かがみの孤城/"`. If you don't want to use split files, make sure there isn't a folder with that name existing, or the program will automatically look for the split files.
 5. Single media file should be in `./<name>/<name>.m4b`. If you have the split audiobook as m4b,mp3, or mp4's you can run `./merge.sh "<full folder path>"`,
  eg `./merge.sh "/mnt/d/Editing/Audiobooks/ｍｅｄｉｕｍ霊媒探偵城塚翡翠"`
-6. If you have the `script.txt` and `./<name>/<name>_splitted/`, you can now run the GPU intense, time intense, and occasionally CPU intense script part. `./run.sh "<full folder path>"` eg `./run.sh "/mnt/d/Editing/Audiobooks/かがみの孤城/"`. This runs each split file individually to get a word level transcript. It then creates a sub format that can be matched to the `script.txt`. Each word level subtitle is merged into a phrase level, and your result should be a `<name>.srt` file that can be watched with `mpv`, showing audio in time with the full book as a subtitle. From there use a texthooker and enjoy.
+6. If you have the `script.txt` and either `./<name>/<name>.m4b` or `./<name>/<name>_splitted/`, you can now run the GPU intense, time intense, and occasionally CPU intense script part. `./run.sh "<full folder path>"` eg `./run.sh "/mnt/d/Editing/Audiobooks/かがみの孤城/"`. This runs each split file individually to get a word level transcript. It then creates a sub format that can be matched to the `script.txt`. Each word level subtitle is merged into a phrase level, and your result should be a `<name>.srt` file that can be watched with `MPV`, showing audio in time with the full book as a subtitle. 
+7. From there, use a [texthooker](https://github.com/Renji-XD/texthooker-ui) with something like [mpv_websocket](https://github.com/kuroahna/mpv_websocket) and enjoy Immersion Reading.
 
 # Split m4b by chapter
 `./split.sh "/mnt/d/Editing/Audiobooks/かがみの孤城/"`
@@ -50,7 +51,7 @@ Primarily I'm using this for syncing audiobooks to their book script. So while y
 
 # Single File
 
-You can also run for a single file. Beware if it's over 1GB/19hr you need as much as 23GB of RAM available.
+You can also run for a single file. Beware if it's over 1GB/19hr you need as much as 8GB of RAM available.
 You need two copies of your file. One in "<full folder path>" and one in `<full folder path>/splitted_<name>`, as described in the How to Use section. The single file will only run if you don't have `<name>_splitted` folder, otherwise we'll assume you want to use the data from there in parts.
 
 `./run.sh "<full folder path>"` eg `./run.sh "$(wslpath -a "D:\Editing\Audiobooks\かがみの孤城\\")"`