Skip to content

Speakr is a personal, self-hosted web application designed for transcribing audio recordings

Python

Speaker modal: stage per-line speaker/text edits until Save; fix one-change-hits-many-lines

Murtaza Nasir1f4f76782425
diff --git a/static/js/app.modular.js b/static/js/app.modular.js
index 5972bb00..72525932 100644
--- a/static/js/app.modular.js
+++ b/static/js/app.modular.js
@@ -2037,6 +2037,12 @@ document.addEventListener('DOMContentLoaded', async () => {
const isNewSpeaker = segment.speakerId !== lastSpeakerId;
processedSimpleSegments.push({
...segment,
+ // The true index into the raw transcription array.
+ // The speaker modal MUST use this (not the v-for
+ // loop index) for change-speaker / edit-text so the
+ // edit hits the exact segment — matches the public
+ // release and avoids v-memo/index drift.
+ _originalIndex: idx,
showSpeaker: isNewSpeaker
});
if (isNewSpeaker) {
diff --git a/static/js/modules/composables/speakers.js b/static/js/modules/composables/speakers.js
index a9a1009f..8d164253 100644
--- a/static/js/modules/composables/speakers.js
+++ b/static/js/modules/composables/speakers.js
@@ -28,6 +28,15 @@ export function useSpeakers(state, utils, processedTranscription) {
// Current speaker highlight state
let currentSpeakerId = null;
+ // Snapshot of the recording's transcription taken when the modal opens.
+ // Per-line speaker/text edits are now STAGED in memory (changeSpeaker /
+ // saveEditedText mutate selectedRecording.transcription for live preview
+ // but don't persist). If the user Cancels with unsaved staged edits, we
+ // restore this snapshot so the cancelled edits don't leak into the
+ // detail view. On a real save the staged flag is cleared first so no
+ // revert happens.
+ let originalTranscriptionSnapshot = null;
+
// Number of speaker colors available in CSS (must match styles.css and app.modular.js)
const SPEAKER_COLOR_COUNT = 16;
@@ -64,6 +73,11 @@ export function useSpeakers(state, utils, processedTranscription) {
// Pause outer audio player to avoid conflicts with modal's player
pauseOuterAudioPlayer();
+ // Snapshot the committed transcription and clear any stale staged
+ // edits so this modal session starts from a clean, saved state.
+ originalTranscriptionSnapshot = selectedRecording.value.transcription || null;
+ editedTranscriptData.value = null;
+
// Clear any existing speaker map data first
speakerMap.value = {};
speakerDisplayMap.value = {};
@@ -177,6 +191,16 @@ export function useSpeakers(state, utils, processedTranscription) {
utils.resetModalAudioState();
}
+ // If the modal is closing with UNSAVED staged per-line edits (the
+ // user hit Cancel/X, not Save), discard them by restoring the
+ // snapshot taken on open. A real save clears editedTranscriptData
+ // before calling this, so saved edits are never reverted.
+ if (editedTranscriptData.value && originalTranscriptionSnapshot != null && selectedRecording.value) {
+ selectedRecording.value.transcription = originalTranscriptionSnapshot;
+ }
+ editedTranscriptData.value = null;
+ originalTranscriptionSnapshot = null;
+
showSpeakerModal.value = false;
showAutoIdDropdown.value = false;
highlightedSpeaker.value = null;
@@ -262,6 +286,10 @@ export function useSpeakers(state, utils, processedTranscription) {
const data = await response.json();
if (!response.ok) throw new Error(data.error || 'Failed to update transcript');
+ // The edits are now persisted on the server — clear the staged
+ // flag BEFORE closing so closeSpeakerModal doesn't mistake this
+ // for a Cancel and revert the just-saved changes.
+ editedTranscriptData.value = null;
closeSpeakerModal();
// If summary regeneration was requested, update status immediately
@@ -1139,23 +1167,20 @@ export function useSpeakers(state, utils, processedTranscription) {
editedText.value = '';
};
- const saveEditedText = async () => {
+ const saveEditedText = () => {
if (editingSegmentIndex.value === null || !selectedRecording.value?.transcription) return;
try {
const transcriptionData = JSON.parse(selectedRecording.value.transcription);
if (transcriptionData && Array.isArray(transcriptionData) && transcriptionData[editingSegmentIndex.value]) {
transcriptionData[editingSegmentIndex.value].sentence = editedText.value;
- editedTranscriptData.value = transcriptionData;
- // Update the recording's transcription temporarily for UI update
+ // Stage in memory only (same as changeSpeaker) — persisted on
+ // "Save Names". No surprise live save.
+ editedTranscriptData.value = transcriptionData;
selectedRecording.value.transcription = JSON.stringify(transcriptionData);
closeEditTextModal();
-
- // Immediately persist the change
- showToast(t('help.savingProgress'), 'fa-spinner fa-spin');
- await saveTranscriptImmediately(transcriptionData);
}
} catch (e) {
console.error('Error saving text:', e);
@@ -1171,23 +1196,32 @@ export function useSpeakers(state, utils, processedTranscription) {
editingSpeakerIndex.value = editingSpeakerIndex.value === segmentIndex ? null : segmentIndex;
};
- const changeSpeaker = async (segmentIndex, newSpeakerId) => {
+ const changeSpeaker = (segmentIndex, newSpeakerId) => {
if (!selectedRecording.value?.transcription) return;
try {
const transcriptionData = JSON.parse(selectedRecording.value.transcription);
if (transcriptionData && Array.isArray(transcriptionData) && transcriptionData[segmentIndex]) {
transcriptionData[segmentIndex].speaker = newSpeakerId;
+
+ // STAGE the change in memory only. It is persisted when the
+ // user clicks "Save Names" (saveSpeakerNames sees
+ // editedTranscriptData and calls saveTranscriptEdits).
+ //
+ // We deliberately do NOT save immediately here. The old
+ // behaviour called saveTranscriptImmediately on every single
+ // per-line change, which (a) saved without the user asking
+ // (the surprise "changes saved" toast) and (b) re-sent the
+ // full speaker_map, which the backend applies by renaming
+ // EVERY segment matching a mapped speaker — so changing one
+ // line silently rewrote all the other lines of that speaker.
editedTranscriptData.value = transcriptionData;
- // Update the recording's transcription temporarily for UI update
+ // Update the recording's transcription in memory so the
+ // modal reflects the change live (still unsaved).
selectedRecording.value.transcription = JSON.stringify(transcriptionData);
editingSpeakerIndex.value = null;
-
- // Immediately persist the change
- showToast(t('help.savingProgress'), 'fa-spinner fa-spin');
- await saveTranscriptImmediately(transcriptionData);
}
} catch (e) {
console.error('Error changing speaker:', e);
diff --git a/templates/modals/speaker-modal.html b/templates/modals/speaker-modal.html
index 5d5c51d1..414c2846 100644
--- a/templates/modals/speaker-modal.html
+++ b/templates/modals/speaker-modal.html
@@ -289,12 +289,12 @@
rendering regressions. v-memo per row stops
Vue from re-rendering unrelated rows when
one row's data changes. -->
- <div v-for="(segment, idx) in processedTranscription.simpleSegments"
- :key="segment.startTime + '-' + idx"
- v-memo="[idx, segment.speakerId, segment.speaker, segment.sentence, highlightedSpeaker === segment.speakerId, editingSpeakerIndex === idx]"
+ <div v-for="segment in processedTranscription.simpleSegments"
+ :key="segment._originalIndex"
+ v-memo="[segment._originalIndex, segment.speakerId, segment.speaker, segment.sentence, highlightedSpeaker === segment.speakerId, editingSpeakerIndex === segment._originalIndex]"
class="speaker-segment group relative"
:data-start-time="segment.startTime"
- :data-segment-index="idx">
+ :data-segment-index="segment._originalIndex">
<div class="flex items-start gap-2">
<span :class="[segment.color, 'speaker-tag', { 'speaker-highlight': highlightedSpeaker === segment.speakerId }]"
:data-speaker-id="segment.speakerId"
@@ -302,15 +302,15 @@
:title="segment.speaker">${ segment.speaker }</span>
<span class="word cursor-pointer hover:bg-[var(--bg-accent)] hover:text-[var(--text-accent)] rounded px-0.5 flex-1 text-xs sm:text-sm leading-relaxed">${ segment.sentence }</span>
<span class="flex items-center gap-0.5 flex-shrink-0 sm:opacity-0 sm:group-hover:opacity-100 transition-opacity">
- <button @click.stop="openSpeakerChangeDropdown(idx)" class="btn btn--icon btn--sm" title="Change speaker">
+ <button @click.stop="openSpeakerChangeDropdown(segment._originalIndex)" class="btn btn--icon btn--sm" title="Change speaker">
<i class="fas fa-user-edit text-[10px]"></i>
</button>
- <button @click.stop="openEditTextModal(idx)" class="btn btn--icon btn--sm" title="Edit text">
+ <button @click.stop="openEditTextModal(segment._originalIndex)" class="btn btn--icon btn--sm" title="Edit text">
<i class="fas fa-edit text-[10px]"></i>
</button>
</span>
</div>
- <div v-if="editingSpeakerIndex === idx"
+ <div v-if="editingSpeakerIndex === segment._originalIndex"
@click.stop
class="surface-raised"
style="margin-top: var(--sp-2); padding: var(--sp-2);">
@@ -318,7 +318,7 @@
<div class="flex flex-wrap gap-1">
<button v-for="speaker in modalSpeakers"
:key="speaker"
- @click="changeSpeaker(idx, speaker)"
+ @click="changeSpeaker(segment._originalIndex, speaker)"
class="px-2 py-1 rounded-md hover:bg-[var(--bg-accent)] hover:text-[var(--text-accent)] active:scale-[0.98] transition-all flex items-center gap-1 text-xs bg-[var(--bg-tertiary)]">
<div :class="speakerMap[speaker].color" class="w-2.5 h-2.5 rounded-full flex-shrink-0"></div>
<span class="truncate max-w-[60px]">${ speakerMap[speaker].name || speaker }</span>