transcription with openai whisperHEADmaster

2 files changed, 70 insertions, 0 deletions

diff --git a/transcribe b/transcribe
new file mode 100755
index 0000000..289bcd2
--- /dev/null
+++ b/transcribe
@@ -0,0 +1,22 @@
+#!/bin/sh
+# Transcribe an audio file using OpenAI/GGML Whisper.cpp.
+
+set -e
+
+model="${XDG_DATA_HOME}/whisper/models/ggml-large-v3-turbo.bin"
+transcribe()
+{
+	echo "Transcribing '$1'" >&2
+	whisper-cli --language English -m "$model" --threads $(nproc) -osrt "$1"
+}
+
+ext="${1##*.}"
+if [ ! "$ext" = "wav" ]; then
+	wav=".${1}.wav"
+	echo "Converting to wav '$1' -> '$wav'" >&2
+	ffmpeg -i "$1" -ar 16k -ac 1 -c:a pcm_s16le "$wav"
+	transcribe "$wav"
+	rm "$wav"
+else
+	transcribe "$1"
+fi
diff --git a/transcribeall b/transcribeall
new file mode 100755
index 0000000..a2de999
--- /dev/null
+++ b/transcribeall
@@ -0,0 +1,48 @@
+#!/bin/sh
+
+set -e
+
+usage="$0 -o <output_dir> <file>..."
+
+# Parse flags
+outdir=""
+while getopts "o:" opt; do
+	case "$opt" in
+		o)
+			outdir="${OPTARG%/}";;
+		*)
+			echo Usage: $usage >&2
+			exit 1;;
+	esac
+done
+shift $((OPTIND - 1))
+if [ -z "$outdir" ]; then
+	echo Usage: $usage >&2
+	exit 1
+fi
+
+# Transcribe each file
+while [ $# -gt 0 ]; do
+	in="$1"
+	base="${1%.*}"
+	out="${outdir}/${base}.srt"
+	wip="${outdir}/.${base}.srt.wip" # work-in-progress
+	shift
+
+	# Check work-in-progress from previous run
+	if [ -e "$wip" ]; then
+		echo "Warning: '$out' in-progress; overwriting." >&2
+		rm -f "$out" "$wip" # remove partial output and restart from scratch
+	fi
+
+	# Transcribe if not done already
+	if [ ! -e "$out" ]; then
+		touch "$wip"
+		echo "'$in' -> '$out'" >&2
+		transcribe "$in" >"$out"
+		rm "$wip"
+	else
+		echo "'$out' already exists; skipping." >&2
+	fi
+done
+echo done >&2