257 lines
7.5 KiB
Bash
257 lines
7.5 KiB
Bash
path_exists() {
|
|
[[ -e "$1" || -L "$1" ]]
|
|
}
|
|
|
|
is_recording() {
|
|
case "${1,,}" in
|
|
*.wav | *.flac | *.mp3 | *.ogg | *.opus | *.aac | *.m4a) return 0 ;;
|
|
*) return 1 ;;
|
|
esac
|
|
}
|
|
|
|
move_no_clobber() {
|
|
local source=$1
|
|
local destination=$2
|
|
|
|
if path_exists "$destination"; then
|
|
return 1
|
|
fi
|
|
|
|
mv -T --no-clobber -- "$source" "$destination"
|
|
! path_exists "$source" && path_exists "$destination"
|
|
}
|
|
|
|
fail_recording() {
|
|
local source=$1
|
|
local failed_dir=$2
|
|
local message=$3
|
|
local details=${4:-}
|
|
local base target error_tmp suffix=0
|
|
|
|
base=$(basename "$source")
|
|
target="$failed_dir/$base"
|
|
rm -f -- "${source}.transcribe-state"
|
|
|
|
while path_exists "$target" || path_exists "${target}.error.txt"; do
|
|
((suffix += 1))
|
|
target="$failed_dir/$base.failed-$suffix"
|
|
done
|
|
|
|
if ! move_no_clobber "$source" "$target"; then
|
|
printf 'Could not quarantine %s: destination collision or move failure\n' "$source" >&2
|
|
return 1
|
|
fi
|
|
|
|
error_tmp=$(mktemp "$failed_dir/.transcription-error.XXXXXX")
|
|
{
|
|
printf '%s\n' "$message"
|
|
if [[ -n "$details" && -s "$details" ]]; then
|
|
printf '\n'
|
|
cat -- "$details"
|
|
fi
|
|
} >"$error_tmp"
|
|
|
|
if ! ln -- "$error_tmp" "${target}.error.txt"; then
|
|
printf 'Could not publish error sidecar for %s\n' "$target" >&2
|
|
rm -f -- "$error_tmp"
|
|
return 1
|
|
fi
|
|
rm -f -- "$error_tmp"
|
|
printf 'Moved failed recording to %s\n' "$target" >&2
|
|
}
|
|
|
|
recover_completed_recording() {
|
|
local source=$1
|
|
local transcript=$2
|
|
local archive=$3
|
|
local state_file="${source}.transcribe-state"
|
|
|
|
if [[ ! -f "$state_file" ]]; then
|
|
return 1
|
|
fi
|
|
|
|
if ! path_exists "$transcript"; then
|
|
rm -f -- "$state_file"
|
|
return 1
|
|
fi
|
|
|
|
if ! move_no_clobber "$source" "$archive"; then
|
|
return 2
|
|
fi
|
|
|
|
rm -f -- "$state_file"
|
|
printf 'Recovered completed transcription and archived %s\n' "$archive"
|
|
return 0
|
|
}
|
|
|
|
process_recording() (
|
|
set -o errexit
|
|
|
|
local source=$1
|
|
local inbox_dir=$2
|
|
local transcript_dir=$3
|
|
local archive_dir=$4
|
|
local failed_dir=$5
|
|
local base stem transcript archive original_path recorded_at transcribed_at
|
|
local work_dir log_file raw_output rendered_output draft state_file recovery_status
|
|
|
|
base=$(basename "$source")
|
|
stem=${base%.*}
|
|
transcript="$transcript_dir/$stem.md"
|
|
archive="$archive_dir/$base"
|
|
original_path="$inbox_dir/$base"
|
|
state_file="${source}.transcribe-state"
|
|
|
|
if recover_completed_recording "$source" "$transcript" "$archive"; then
|
|
return 0
|
|
else
|
|
recovery_status=$?
|
|
fi
|
|
if [[ $recovery_status -eq 2 ]]; then
|
|
fail_recording "$source" "$failed_dir" "Archive collision after transcription: $archive"
|
|
return 1
|
|
fi
|
|
|
|
if path_exists "$transcript"; then
|
|
fail_recording "$source" "$failed_dir" "Transcript already exists: $transcript"
|
|
return 1
|
|
fi
|
|
if path_exists "$archive"; then
|
|
fail_recording "$source" "$failed_dir" "Archived recording already exists: $archive"
|
|
return 1
|
|
fi
|
|
|
|
recorded_at=$(date --date="@$(stat -c %Y -- "$source")" --iso-8601=seconds)
|
|
transcribed_at=$(date --utc --iso-8601=seconds)
|
|
work_dir=$(mktemp -d "$runtime_dir/work.XXXXXX")
|
|
trap 'rm -r -- "$work_dir"' EXIT
|
|
log_file="$work_dir/transcription.log"
|
|
raw_output="$work_dir/transcript"
|
|
rendered_output="$work_dir/transcript.md"
|
|
|
|
if ! ffmpeg \
|
|
-nostdin \
|
|
-hide_banner \
|
|
-loglevel error \
|
|
-i "$source" \
|
|
-vn \
|
|
-ac 1 \
|
|
-ar 16000 \
|
|
-c:a pcm_s16le \
|
|
"$work_dir/audio.wav" >"$log_file" 2>&1; then
|
|
fail_recording "$source" "$failed_dir" "FFmpeg conversion failed" "$log_file"
|
|
return 1
|
|
fi
|
|
|
|
if ! whisper-cli \
|
|
--model "$TRANSCRIBE_MODEL" \
|
|
--file "$work_dir/audio.wav" \
|
|
--language auto \
|
|
--output-json-full \
|
|
--output-file "$raw_output" >>"$log_file" 2>&1 || [[ ! -s "${raw_output}.json" ]]; then
|
|
fail_recording "$source" "$failed_dir" "Whisper transcription failed" "$log_file"
|
|
return 1
|
|
fi
|
|
|
|
if ! diarize-transcript \
|
|
--audio "$work_dir/audio.wav" \
|
|
--model "$TRANSCRIBE_DIARIZATION_MODEL" \
|
|
--transcript "${raw_output}.json" \
|
|
--output "$rendered_output" >>"$log_file" 2>&1 || [[ ! -s "$rendered_output" ]]; then
|
|
fail_recording "$source" "$failed_dir" "Speaker diarization failed" "$log_file"
|
|
return 1
|
|
fi
|
|
|
|
draft=$(mktemp "$transcript_dir/.transcript.XXXXXX")
|
|
{
|
|
printf '# %s\n\n' "$stem"
|
|
printf -- '- Original recording: %s\n' "$original_path"
|
|
printf -- '- Recording timestamp: %s\n' "$recorded_at"
|
|
printf -- '- Transcription timestamp: %s\n' "$transcribed_at"
|
|
printf -- '- Engine: whisper.cpp %s (Vulkan)\n' "$TRANSCRIBE_ENGINE_VERSION"
|
|
printf -- '- Model: %s\n' "$TRANSCRIBE_MODEL_NAME"
|
|
printf -- '- Model SHA-256: %s\n' "$TRANSCRIBE_MODEL_SHA256"
|
|
printf -- '- Diarization engine: pyannote.audio %s (CPU)\n' "$TRANSCRIBE_DIARIZATION_ENGINE_VERSION"
|
|
printf -- '- Diarization model: %s\n' "$TRANSCRIBE_DIARIZATION_MODEL_NAME"
|
|
printf -- '- Diarization model revision: %s\n\n' "$TRANSCRIBE_DIARIZATION_MODEL_REVISION"
|
|
cat -- "$rendered_output"
|
|
} >"$draft"
|
|
chmod 0644 "$draft"
|
|
|
|
touch -- "$state_file"
|
|
|
|
if ! ln -- "$draft" "$transcript"; then
|
|
rm -f -- "$draft" "$state_file"
|
|
fail_recording "$source" "$failed_dir" "Transcript appeared while transcription was running: $transcript" "$log_file"
|
|
return 1
|
|
fi
|
|
rm -f -- "$draft"
|
|
|
|
if ! move_no_clobber "$source" "$archive"; then
|
|
printf 'Could not archive %s; leaving recoverable state in processing\n' "$source" >&2
|
|
return 1
|
|
fi
|
|
|
|
rm -f -- "$state_file"
|
|
printf 'Transcribed %s to %s\n' "$archive" "$transcript"
|
|
)
|
|
|
|
process_scope() {
|
|
local recordings_dir=$1
|
|
local transcript_dir=$2
|
|
local inbox_dir="$recordings_dir/inbox"
|
|
local processing_dir="$recordings_dir/processing"
|
|
local archive_dir="$recordings_dir/archive"
|
|
local failed_dir="$recordings_dir/failed"
|
|
local source processing_target recording_status scope_status=0
|
|
|
|
if ! mkdir -p "$inbox_dir" "$processing_dir" "$archive_dir" "$failed_dir" "$transcript_dir"; then
|
|
return 1
|
|
fi
|
|
|
|
while IFS= read -r -d '' source; do
|
|
if is_recording "$source"; then
|
|
process_recording "$source" "$inbox_dir" "$transcript_dir" "$archive_dir" "$failed_dir"
|
|
recording_status=$?
|
|
if [[ $recording_status -ne 0 ]]; then
|
|
scope_status=1
|
|
fi
|
|
fi
|
|
done < <(find "$processing_dir" -maxdepth 1 -type f -print0 | sort -z)
|
|
|
|
while IFS= read -r -d '' source; do
|
|
if ! is_recording "$source"; then
|
|
continue
|
|
fi
|
|
|
|
processing_target="$processing_dir/$(basename "$source")"
|
|
if ! move_no_clobber "$source" "$processing_target"; then
|
|
fail_recording "$source" "$failed_dir" "Processing filename already exists: $processing_target"
|
|
scope_status=1
|
|
continue
|
|
fi
|
|
process_recording "$processing_target" "$inbox_dir" "$transcript_dir" "$archive_dir" "$failed_dir"
|
|
recording_status=$?
|
|
if [[ $recording_status -ne 0 ]]; then
|
|
scope_status=1
|
|
fi
|
|
done < <(find "$inbox_dir" -maxdepth 1 -type f -print0 | sort -z)
|
|
|
|
return "$scope_status"
|
|
}
|
|
|
|
runtime_dir="${XDG_RUNTIME_DIR:?}/transcribe-recordings"
|
|
mkdir -p "$runtime_dir"
|
|
|
|
if [[ ! -f "$TRANSCRIBE_DIARIZATION_MODEL/.complete" || ! -f "$TRANSCRIBE_DIARIZATION_MODEL/config.yaml" ]]; then
|
|
printf 'Speaker diarization model is not installed; run prepare-transcription-diarization-model\n' >&2
|
|
exit 1
|
|
fi
|
|
|
|
set +o errexit
|
|
process_scope "$TRANSCRIBE_SYNC_ROOT/recordings" "$TRANSCRIBE_SYNC_ROOT/notes/transcripts"
|
|
personal_status=$?
|
|
process_scope "$TRANSCRIBE_SYNC_ROOT/work/recordings" "$TRANSCRIBE_SYNC_ROOT/work/notes/transcripts"
|
|
work_status=$?
|
|
exit "$((personal_status != 0 || work_status != 0))"
|