Both sides previous revision Previous revision | |
events:divocptt [2020/10/04 19:25] – [Special Post-Production Sauce due to Separate Translation] stb | events:divocptt [2020/10/04 21:31] (current) – danimo |
---|
3. Make sure you have the translator audio for the respective time. See https://wiki.c3lingo.org/translation-recordings | 3. Make sure you have the translator audio for the respective time. See https://wiki.c3lingo.org/translation-recordings |
4. Use ffmpeg to cut the mp4 to the required length without re-encoding: ''ffmpeg -i muxed.mp4 -ss <start> -to <end> -c copy id-<hd>-orig.mp4'', where ''<start>'' and ''<end>'' are timestamps and <id> is the talk's id in the tracker. Check the final length (ffmpeg output or run ffprobe). Note that this method can only split at keyfames, so the final result might be a bit off. If that's not tolerable, you will have to do a reencode (instead of ''-c copy'', use ''-c:a copy -c:v libx264 -bufsize:0 8192k -minrate:0 100k -maxrate:0 2000k -crf:0 18 -profile:0 main''). | 4. Use ffmpeg to cut the mp4 to the required length without re-encoding: ''ffmpeg -i muxed.mp4 -ss <start> -to <end> -c copy id-<hd>-orig.mp4'', where ''<start>'' and ''<end>'' are timestamps and <id> is the talk's id in the tracker. Check the final length (ffmpeg output or run ffprobe). Note that this method can only split at keyfames, so the final result might be a bit off. If that's not tolerable, you will have to do a reencode (instead of ''-c copy'', use ''-c:a copy -c:v libx264 -bufsize:0 8192k -minrate:0 100k -maxrate:0 2000k -crf:0 18 -profile:0 main''). |
5. Open the clip in kdenlive or shotcut. open the translation right along with it in a separate audio track. and align the translation so it fits with the original (delay it by a few secs so it feels natural, live translations are never instant). Fit the audio exactly with the video in terms of length, and mute the video's audio. Next, click Render/Export and make sure to only export the audio, as aac/m4a. name the file ''<id>-hd-tl.mp4''. (This can also be done with pure ffmpeg if you feel confident enough) | 5. Open the clip in kdenlive or shotcut. open the translation right along with it in a separate audio track. and align the translation so it fits with the original (delay it by a few secs so it feels natural, live translations are never instant). Fit the audio exactly with the video in terms of length, and mute the video's audio. Next, click Render/Export and make sure to only export the audio, as aac/m4a. name the file ''<id>-hd-tl.m4a''. (This can also be done with pure ffmpeg if you feel confident enough) |
6. Merge translation to the original video: ''ffmpeg -i <hd>-orig.mp4 -i <id>-hd-tl.mp4 -c copy -map 0:v -map 0:a -map 1:a -shortest <id>-hd.mp4'' | 6. Merge translation to the original video: ''ffmpeg -i <hd>-orig.mp4 -i <id>-hd-tl.m4a -c copy -map 0:v -map 0:a -map 1:a -shortest <id>-hd.mp4'' |
7. Verify with ffprobe that there now are one video and two audio tracks on the new file. | 7. Verify with ffprobe that there now are one video and two audio tracks on the new file. |
8. Copy the new file to ''releasing.c3voc.de:/video/tmp/divoc_ptt/repair'' | 8. Copy the new file to ''releasing.c3voc.de:/video/tmp/divoc_ptt/repair'' |