Add scripts for audio transcription using whisper.cpp

bin/transcribe-video-batch

@@ -0,0 +1,65 @@
+#!/usr/bin/env bash
+
+# I was originally just using three calls to transcribe-video-with-model but I want to reuse the same audio input, so this
+# is mostly a copy pasta of that file.
+
+if which tput >/dev/null 2>&1; then
+    ncolors=$(tput colors)
+fi
+if [ -t 1 ] && [ -n "$ncolors" ] && [ "$ncolors" -ge 8 ]; then
+    RED="$(tput setaf 1)"
+    GREEN="$(tput setaf 2)"
+    YELLOW="$(tput setaf 3)"
+    BLUE="$(tput setaf 4)"
+    MAGENTA="$(tput setaf 5)"
+    CYAN="$(tput setaf 6)"
+    BOLD="$(tput bold)"
+    NORMAL="$(tput sgr0)"
+else
+    RED=""
+    GREEN=""
+    YELLOW=""
+    BLUE=""
+    MAGENTA=""
+    CYAN=""
+    BOLD=""
+    NORMAL=""
+fi
+
+input_video="$1"
+output_name_without_ext="$2"
+threads=$3
+
+if [[ $input_video == "" || $output_name_without_ext == "" ]]; then
+    printf "${BOLD}${RED}Usage: $0 <input.wav> <output name without extension> <optional: thread count>${NORMAL}\n"
+    exit 1
+fi
+
+wav_name="${output_name_without_ext}_audio_${RANDOM}.wav"
+
+extract-16bit-wav-from-video "$input_video" "$wav_name"
+if [[ $? == 1 ]]; then exit 1; fi
+
+#
+# Tiny model first to have something quickly banged out. base and small have similar output quality. Neither are perfect.
+#
+
+transcribe-audio "$wav_name" "$output_name_without_ext" "tiny" $threads
+if [[ $? == 1 ]]; then
+    printf "${RED}${BOLD}Saving the audio file \"$wav_name\" in case you want to reuse it for debugging.\n${NORMAL}"
+    exit 1
+fi
+
+transcribe-audio "$wav_name" "$output_name_without_ext" "base" $threads
+if [[ $? == 1 ]]; then
+    printf "${RED}${BOLD}Saving the audio file \"$wav_name\" in case you want to reuse it for debugging.\n${NORMAL}"
+    exit 1
+fi
+
+transcribe-audio "$wav_name" "$output_name_without_ext" "small" $threads
+if [[ $? == 1 ]]; then
+    printf "${RED}${BOLD}Saving the audio file \"$wav_name\" in case you want to reuse it for debugging.\n${NORMAL}"
+    exit 1
+fi
+
+rm "$wav_name"