transcribe-matrix-live

commit f376b4009c3f3cda7927801ad0a44f5e4b91196c
parent 402974eeb783c1f198c8c924f17f83f77dd49c8f
Author: MTRNord <mtrnord1@gmail.com>
Date:   Wed,  3 Jan 2024 22:35:16 +0100

uhhhh

Diffstat:
M
transcribe.sh
 | 
21
+++++++++++----------

1 file changed, 11 insertions(+), 10 deletions(-)
diff --git a/transcribe.sh b/transcribe.sh
@@ -62,31 +62,31 @@ do
 done
 
 echo "normalize"
-files_already_done=($(find . -wholename "./playlist_normalized/*.wav" -type f | tr '\n' ' ' | sed 's/playlist_normalized/playlist/g'))
+files_already_done=($(find . -wholename "./playlist_normalized/*.wav" -type f | tr '\n' ' ' | sed -e 's/playlist_normalized/playlist/g'))
 files=$(find . -wholename "./playlist/*.wav" -type f | tr '\n' ' ')
 for already_done_file in "${files_already_done[@]}"
 do
-    files=$(echo "${files}" | sed "s@${already_done_file}@@g")
+    files=$(echo "${files}" | sed -e "s@${already_done_file}@@g")
 done
-files_out=$(echo "${files}" | sed 's/playlist/playlist_normalized/g')
+files_out=$(echo "${files}" | sed -e 's/playlist/playlist_normalized/g')
 
 [[ $files = *[!\ ]* ]] && python3 -m ffmpeg_normalize $files -p -o $files_out -ar 16000
 
 ## This logic is used to not process stuff twice
-#files_already_done=($(find . -wholename "./output/*.vtt" -type f | tr '\n' ' ' | sed 's/output/playlist_normalized/g' | sed 's/vtt/wav/g'))
-files_already_done=($(find . -wholename "./output/*.txt" -type f | tr '\n' ' ' | sed 's/output/playlist_normalized/g'| sed 's/txt/wav/g'))
+#files_already_done=($(find . -wholename "./output/*.vtt" -type f | tr '\n' ' ' | sed -e 's/output/playlist_normalized/g' | sed -e 's/vtt/wav/g'))
+files_already_done=($(find . -wholename "./output/*.txt" -type f | tr '\n' ' ' | sed -e 's/output/playlist_normalized/g'| sed -e 's/txt/wav/g'))
 files=$(find . -wholename "./playlist_normalized/*.wav" -type f | tr '\n' ' ')
 for already_done_file in "${files_already_done[@]}"
 do
-    files=$(echo "${files}" | sed "s@${already_done_file}@@g")
+    files=$(echo "${files}" | sed -e "s@${already_done_file}@@g")
 done
-out_files=$(echo "${files}" | sed 's/playlist_normalized/output/g' | sed 's/.wav//g')
+out_files=$(echo "${files}" | sed -e 's/playlist_normalized/output/g' | sed -e 's/.wav//g')
 	
 echo "starting whisper"
 
 if [[ $files = *[!\ ]* ]]; then
     #if ! nice -n 18 ./main -m "models/ggml-${MODEL}.bin" -t "$THREADS" -l en -ovtt -pc "${files}.wav"; #>/dev/null  2>/dev/null;
-    if ! nice -n 18 ./main -m "models/ggml-${MODEL}.bin" -t "$THREADS" -l en -otxt -pc --file $files --output-file $out_files -et 3.0; #>/dev/null  2>/dev/null;
+    if ! nice -n 18 ./main -m "models/ggml-${MODEL}.bin" -t "$THREADS" -l en -otxt -ovtt -pc --file $files --output-file $out_files -et 3.0; #>/dev/null  2>/dev/null;
         then
             echo "error transcribing"
     fi
@@ -98,11 +98,12 @@ if [[ $files = *[!\ ]* ]]; then
         FILENAME=$(basename "${next_file}" ".wav")
         # Removes some junk from hallucination
         sed -i -e 's/ Subtitles by the Amara.org community//g' "./output/${FILENAME}.txt"
+        sed -i -e 's/Subtitles by the Amara.org community//g' "./output/${FILENAME}.vtt"
 
-        rm "./playlist/${FILENAME}.wav"
+        mv "./playlist/${FILENAME}.wav" "./playlist_bak/${FILENAME}.wav"
         rm "./playlist_normalized/${FILENAME}.wav"
         #rm "./${FILENAME}.vtt"
-        #mv "./${FILENAME}.wav.vtt" "./output/${FILENAME}.vtt"
+        mv "./${FILENAME}.wav.vtt" "./output/${FILENAME}.vtt"
         mv "./${FILENAME}.wav.txt" "./output/${FILENAME}.txt"
     done
 fi