Skip to content

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

Python

fixed json transcripts

Murtaza Nasir7abdf7fbb6e0
diff --git a/app.py b/app.py
index 805a06ff..114877c2 100644
--- a/app.py
+++ b/app.py
@@ -44,6 +44,26 @@ def md_to_html(text):
])
return html
+def format_transcription_for_llm(transcription_text):
+ """
+ Formats transcription for LLM. If it's our simplified JSON, convert it to plain text.
+ Otherwise, return as is.
+ """
+ try:
+ transcription_data = json.loads(transcription_text)
+ if isinstance(transcription_data, list):
+ # It's our simplified JSON format
+ formatted_lines = []
+ for segment in transcription_data:
+ speaker = segment.get('speaker', 'Unknown Speaker')
+ sentence = segment.get('sentence', '')
+ formatted_lines.append(f"[{speaker}]: {sentence}")
+ return "\n".join(formatted_lines)
+ except (json.JSONDecodeError, TypeError):
+ # Not a JSON, or not the format we expect, so return as is.
+ pass
+ return transcription_text
+
app = Flask(__name__)
# Use environment variables or default paths for Docker compatibility
app.config['SQLALCHEMY_DATABASE_URI'] = os.environ.get('SQLALCHEMY_DATABASE_URI', 'sqlite:////data/instance/transcriptions.db')
@@ -435,9 +455,23 @@ def transcribe_audio_asr(app_context, recording_id, filepath, original_filename,
with httpx.Client() as client:
response = client.post(url, params=params, files=files, timeout=None)
response.raise_for_status()
- # The ASR service now returns JSON as a string.
- # We will store it directly. The frontend will parse it.
- recording.transcription = response.text
+
+ # Parse the JSON response from ASR
+ asr_response_data = response.json()
+
+ # Simplify the JSON data
+ simplified_segments = []
+ if 'segments' in asr_response_data and isinstance(asr_response_data['segments'], list):
+ for segment in asr_response_data['segments']:
+ simplified_segments.append({
+ 'speaker': segment.get('speaker'),
+ 'sentence': segment.get('text', '').strip(),
+ 'start_time': segment.get('start'),
+ 'end_time': segment.get('end')
+ })
+
+ # Store the simplified JSON as a string
+ recording.transcription = json.dumps(simplified_segments)
app.logger.info(f"ASR transcription completed for recording {recording_id}.")
generate_summary_task(app_context, recording_id)
@@ -683,15 +717,16 @@ def update_speakers(recording_id):
is_json = False
try:
transcription_data = json.loads(transcription_text)
- is_json = isinstance(transcription_data, dict) and 'segments' in transcription_data
+ # Updated check for our new simplified JSON format (a list of segment objects)
+ is_json = isinstance(transcription_data, list)
except (json.JSONDecodeError, TypeError):
is_json = False
speaker_names_used = []
if is_json:
- # Handle JSON transcript
- for segment in transcription_data.get('segments', []):
+ # Handle new simplified JSON transcript (list of segments)
+ for segment in transcription_data:
original_speaker_label = segment.get('speaker')
if original_speaker_label in speaker_map:
new_name_info = speaker_map[original_speaker_label]
@@ -707,9 +742,8 @@ def update_speakers(recording_id):
recording.transcription = json.dumps(transcription_data)
# Update participants from the final list of speakers in the JSON
- if transcription_data.get('segments'):
- final_speakers = set(seg.get('speaker') for seg in transcription_data['segments'] if seg.get('speaker'))
- recording.participants = ', '.join(sorted(list(final_speakers)))
+ final_speakers = set(seg.get('speaker') for seg in transcription_data if seg.get('speaker'))
+ recording.participants = ', '.join(sorted(list(final_speakers)))
else:
# Handle plain text transcript
@@ -761,8 +795,11 @@ def identify_speakers_from_text(transcription):
if not TEXT_MODEL_API_KEY:
raise ValueError("TEXT_MODEL_API_KEY not configured.")
+ # The transcription passed here could be JSON, so we format it.
+ formatted_transcription = format_transcription_for_llm(transcription)
+
# Extract existing speaker labels (e.g., SPEAKER_00, SPEAKER_01) in order of appearance
- all_labels = re.findall(r'\[(SPEAKER_\d+)\]', transcription)
+ all_labels = re.findall(r'\[(SPEAKER_\d+)\]', formatted_transcription)
seen = set()
speaker_labels = [x for x in all_labels if not (x in seen or seen.add(x))]
@@ -773,7 +810,7 @@ def identify_speakers_from_text(transcription):
Transcription:
---
-{transcription[:28000]}
+{formatted_transcription[:28000]}
---
Respond with a single JSON object where keys are the speaker labels (e.g., "SPEAKER_00") and values are the identified full names. If a name cannot be determined, use the value "Unknown".
@@ -818,6 +855,9 @@ def identify_unidentified_speakers_from_text(transcription, unidentified_speaker
if not TEXT_MODEL_API_KEY:
raise ValueError("TEXT_MODEL_API_KEY not configured.")
+ # The transcription passed here could be JSON, so we format it.
+ formatted_transcription = format_transcription_for_llm(transcription)
+
if not unidentified_speakers:
return {}
@@ -827,7 +867,7 @@ Note: Only identify the speakers listed above. Other speakers in the transcripti
Transcription:
---
-{transcription[:28000]}
+{formatted_transcription[:28000]}
---
Respond with a single JSON object where keys are the unidentified speaker labels (e.g., "SPEAKER_01") and values are the identified full names. If a name cannot be determined, use the value "Unknown".
@@ -887,7 +927,8 @@ def auto_identify_speakers(recording_id):
current_speaker_map = data.get('current_speaker_map', {})
# Extract all speaker labels from transcription
- all_labels = re.findall(r'\[(SPEAKER_\d+)\]', recording.transcription)
+ formatted_transcription = format_transcription_for_llm(recording.transcription)
+ all_labels = re.findall(r'\[(SPEAKER_\d+)\]', formatted_transcription)
seen = set()
speaker_labels = [x for x in all_labels if not (x in seen or seen.add(x))]
@@ -905,7 +946,7 @@ def auto_identify_speakers(recording_id):
return jsonify({'success': True, 'speaker_map': {}, 'message': 'All speakers are already identified'})
# Call the helper function with only unidentified speakers
- speaker_map = identify_unidentified_speakers_from_text(recording.transcription, unidentified_speakers)
+ speaker_map = identify_unidentified_speakers_from_text(formatted_transcription, unidentified_speakers)
return jsonify({'success': True, 'speaker_map': speaker_map})
@@ -959,6 +1000,8 @@ def chat_with_transcription():
user_title = current_user.job_title if current_user.is_authenticated and current_user.job_title else "a professional"
user_company = current_user.company if current_user.is_authenticated and current_user.company else "their organization"
+ formatted_transcription = format_transcription_for_llm(recording.transcription)
+
system_prompt = f"""You are a professional meeting and audio transcription analyst assisting {user_name}, who is a(n) {user_title} at {user_company}. {language_instruction} Analyze the following meeting information and respond to the specific request.
Following are the meeting participants and their roles:
@@ -966,7 +1009,7 @@ Following are the meeting participants and their roles:
Following is the meeting transcript:
<<start transcript>>
-{recording.transcription or "No transcript available."}
+{formatted_transcription or "No transcript available."}
<<end transcript>>
Additional context and notes about the meeting:
diff --git a/static/js/app.js b/static/js/app.js
index 40197819..2ae75c36 100644
--- a/static/js/app.js
+++ b/static/js/app.js
@@ -184,16 +184,17 @@ document.addEventListener('DOMContentLoaded', () => {
transcriptionData = null;
}
- if (transcriptionData && transcriptionData.segments) {
+ // Updated to handle new simplified JSON format (array of segments)
+ if (transcriptionData && Array.isArray(transcriptionData)) {
// JSON format - extract speakers from segments
const speakers = new Set();
- transcriptionData.segments.forEach(segment => {
+ transcriptionData.forEach(segment => {
if (segment.speaker) {
speakers.add(segment.speaker);
}
});
- return Array.from(speakers);
- } else {
+ return Array.from(speakers).sort(); // Sort for consistent color mapping
+ } else if (typeof transcription === 'string') {
// Plain text format - use regex to find speaker patterns
const speakerRegex = /\[([^\]]+)\]:/g;
const speakers = new Set();
@@ -201,8 +202,9 @@ document.addEventListener('DOMContentLoaded', () => {
while ((match = speakerRegex.exec(transcription)) !== null) {
speakers.add(match[1]);
}
- return Array.from(speakers);
+ return Array.from(speakers).sort();
}
+ return [];
});
const processedTranscription = computed(() => {
@@ -219,28 +221,23 @@ document.addEventListener('DOMContentLoaded', () => {
transcriptionData = null;
}
- if (transcriptionData && transcriptionData.segments) {
- // JSON processing logic
+ // Updated to handle new simplified JSON format (array of segments)
+ if (transcriptionData && Array.isArray(transcriptionData)) {
const speakerColors = {};
- const speakerList = Array.from(new Set(transcriptionData.segments.map(seg => seg.speaker)));
+ const speakerList = identifiedSpeakers.value; // Use computed property which is already sorted
speakerList.forEach((speaker, index) => {
speakerColors[speaker] = `speaker-color-${(index % 8) + 1}`;
});
- const simpleSegments = transcriptionData.segments.map(segment => ({
+ const simpleSegments = transcriptionData.map(segment => ({
+ speakerId: segment.speaker,
speaker: (speakerMap.value[segment.speaker]?.name || segment.speaker),
- text: segment.text,
- words: segment.words.map((word, index) => ({
- ...word,
- // Add a click handler or data attribute for seeking
- 'data-start-time': word.start,
- // Add space after word unless it's the last word in the segment
- displayWord: word.word + (index < segment.words.length - 1 ? ' ' : '')
- })),
+ sentence: segment.sentence,
+ startTime: segment.start_time,
+ endTime: segment.end_time,
color: speakerColors[segment.speaker] || 'speaker-color-1'
}));
- // Process simple segments to add showSpeaker flag for proper display
const processedSimpleSegments = [];
let lastSpeaker = null;
simpleSegments.forEach(segment => {
@@ -258,14 +255,14 @@ document.addEventListener('DOMContentLoaded', () => {
bubbleRows.push({
speaker: segment.speaker,
color: segment.color,
- isMe: segment.speaker && segment.speaker.toLowerCase().includes('me'),
+ isMe: segment.speaker && (typeof segment.speaker === 'string') && segment.speaker.toLowerCase().includes('me'),
bubbles: []
});
lastBubbleSpeaker = segment.speaker;
}
bubbleRows[bubbleRows.length - 1].bubbles.push({
- text: segment.text,
- words: segment.words,
+ sentence: segment.sentence,
+ startTime: segment.startTime,
color: segment.color
});
});
@@ -322,7 +319,7 @@ document.addEventListener('DOMContentLoaded', () => {
if (currentSpeaker && currentText.trim()) {
segments.push({
speaker: currentSpeaker,
- text: currentText.trim(),
+ sentence: currentText.trim(),
color: speakerColors[currentSpeaker] || 'speaker-color-1'
});
}
@@ -333,7 +330,7 @@ document.addEventListener('DOMContentLoaded', () => {
} else if (!currentSpeaker && line.trim()) {
segments.push({
speaker: null,
- text: line.trim(),
+ sentence: line.trim(),
color: 'speaker-color-1'
});
}
@@ -342,7 +339,7 @@ document.addEventListener('DOMContentLoaded', () => {
if (currentSpeaker && currentText.trim()) {
segments.push({
speaker: currentSpeaker,
- text: currentText.trim(),
+ sentence: currentText.trim(),
color: speakerColors[currentSpeaker] || 'speaker-color-1'
});
}
@@ -352,7 +349,9 @@ document.addEventListener('DOMContentLoaded', () => {
segments.forEach(segment => {
simpleSegments.push({
...segment,
- showSpeaker: segment.speaker !== lastSpeaker
+ showSpeaker: segment.speaker !== lastSpeaker,
+ // Ensure sentence property exists for plain text
+ sentence: segment.sentence || segment.text
});
lastSpeaker = segment.speaker;
});
@@ -370,7 +369,7 @@ document.addEventListener('DOMContentLoaded', () => {
};
}
currentRow.bubbles.push({
- text: segment.text,
+ sentence: segment.sentence,
color: segment.color
});
});
@@ -2092,26 +2091,40 @@ document.addEventListener('DOMContentLoaded', () => {
return;
}
- // Get the button that was clicked
const button = event.currentTarget;
+ let textToCopy = '';
- // Show visual feedback on button
+ try {
+ const transcriptionData = JSON.parse(selectedRecording.value.transcription);
+ if (Array.isArray(transcriptionData)) {
+ // It's our simplified JSON, format it for copying
+ textToCopy = transcriptionData.map(segment => {
+ const speakerName = speakerMap.value[segment.speaker]?.name || segment.speaker;
+ return `[${speakerName}]: ${segment.sentence}`;
+ }).join('\n');
+ } else {
+ // It's some other JSON or plain text, copy as is
+ textToCopy = selectedRecording.value.transcription;
+ }
+ } catch (e) {
+ // Not JSON, so it's plain text
+ textToCopy = selectedRecording.value.transcription;
+ }
+
animateCopyButton(button);
if (navigator.clipboard && window.isSecureContext) {
- // Use clipboard API if available (secure context)
- navigator.clipboard.writeText(selectedRecording.value.transcription)
+ navigator.clipboard.writeText(textToCopy)
.then(() => {
showToast('Transcription copied to clipboard!');
})
.catch(err => {
console.error('Copy failed:', err);
showToast('Failed to copy: ' + err.message, 'fa-exclamation-circle');
- fallbackCopyTextToClipboard(selectedRecording.value.transcription);
+ fallbackCopyTextToClipboard(textToCopy);
});
} else {
- // Fallback for non-secure contexts
- fallbackCopyTextToClipboard(selectedRecording.value.transcription);
+ fallbackCopyTextToClipboard(textToCopy);
}
};
@@ -2181,14 +2194,13 @@ document.addEventListener('DOMContentLoaded', () => {
};
const seekAudioFromEvent = (event) => {
- const wordSpan = event.target.closest('[data-start-time]');
- if (!wordSpan) return;
+ const segmentElement = event.target.closest('[data-start-time]');
+ if (!segmentElement) return;
- const time = parseFloat(wordSpan.dataset.startTime);
+ const time = parseFloat(segmentElement.dataset.startTime);
if (isNaN(time)) return;
// Determine context by checking if we're inside the speaker modal
- // The speaker modal has a specific structure with showSpeakerModal condition
const isInSpeakerModal = event.target.closest('.speaker-modal-transcript') !== null;
const context = isInSpeakerModal ? 'modal' : 'main';
diff --git a/templates/index.html b/templates/index.html
index 318d91c5..0013b8d8 100644
--- a/templates/index.html
+++ b/templates/index.html
@@ -323,20 +323,16 @@
</div>
<!-- Transcript Section -->
- <div class="flex-grow overflow-y-auto custom-scrollbar speaker-modal-transcript">
+ <div class="flex-grow overflow-y-auto custom-scrollbar speaker-modal-transcript" @click="seekAudioFromEvent">
<div v-if="processedTranscription.isJson">
- <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment mb-2">
- <div v-if="segment.showSpeaker"
- :class="[segment.color, 'speaker-tablet', 'font-semibold', 'mb-1', { 'speaker-highlight': highlightedSpeaker === segment.speaker }]"
- :data-speaker-id="segment.speaker">
- ${ segment.speaker }
- </div>
- <div class="speaker-text" @click="seekAudioFromEvent">
- <span v-for="word in segment.words" :key="word.start" class="word cursor-pointer hover:bg-[var(--bg-accent)] hover:text-[var(--text-accent)] rounded px-0.5" :data-start-time="word.start">${ word.displayWord }</span>
- </div>
+ <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment mb-2" :data-start-time="segment.startTime">
+ <p>
+ <span :class="[segment.color, 'speaker-tag', { 'speaker-highlight': highlightedSpeaker === segment.speakerId }]" :data-speaker-id="segment.speakerId">[${ segment.speaker }]</span>:
+ <span class="word cursor-pointer hover:bg-[var(--bg-accent)] hover:text-[var(--text-accent)] rounded px-0.5">${ segment.sentence }</span>
+ </p>
</div>
</div>
- <div v-else v-html="highlightedTranscript" @click="seekAudioFromEvent"></div>
+ <div v-else v-html="highlightedTranscript"></div>
</div>
</div>
</div>
@@ -535,26 +531,24 @@
</div>
<!-- Transcription Content -->
- <div v-if="processedTranscription.hasDialogue">
+ <div v-if="processedTranscription.hasDialogue" @click="seekAudioFromEvent">
<!-- Simple View -->
- <div v-if="transcriptionViewMode === 'simple'" class="transcription-simple-view" @click="seekAudioFromEvent">
- <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment">
+ <div v-if="transcriptionViewMode === 'simple'" class="transcription-simple-view">
+ <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment" :data-start-time="segment.startTime">
<div v-if="segment.showSpeaker" :class="[segment.color, 'speaker-tablet']">${ segment.speaker }</div>
<div class="speaker-text">
- <span v-if="processedTranscription.isJson" v-for="word in segment.words" :key="word.start" class="word" :data-start-time="word.start">${ word.displayWord }</span>
- <span v-else>${ segment.text }</span>
+ <span class="word">${ segment.sentence }</span>
</div>
</div>
</div>
<!-- Bubble View -->
- <div v-else class="transcription-with-speakers" @click="seekAudioFromEvent">
+ <div v-else class="transcription-with-speakers">
<div v-for="(row, index) in processedTranscription.bubbleRows" :key="index"
:class="['bubble-row', { 'speaker-me': row.isMe }]">
<div v-for="(bubble, bubbleIndex) in row.bubbles" :key="bubbleIndex"
- :class="[row.color, 'speaker-bubble', { 'speaker-me': row.isMe }]">
+ :class="[row.color, 'speaker-bubble', { 'speaker-me': row.isMe }]" :data-start-time="bubble.startTime">
<div class="speaker-bubble-content">
- <span v-if="processedTranscription.isJson" v-for="word in bubble.words" :key="word.start" class="word" :data-start-time="word.start">${ word.displayWord }</span>
- <span v-else>${ bubble.text }</span>
+ <span class="word">${ bubble.sentence }</span>
</div>
</div>
</div>
@@ -987,11 +981,10 @@
<div v-if="processedTranscription.hasDialogue" @click="seekAudioFromEvent">
<!-- Simple View -->
<div v-if="transcriptionViewMode === 'simple'" class="transcription-simple-view">
- <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment">
+ <div v-for="(segment, index) in processedTranscription.simpleSegments" :key="index" class="speaker-segment" :data-start-time="segment.startTime">
<div v-if="segment.showSpeaker" :class="[segment.color, 'speaker-tablet']">${ segment.speaker }</div>
<div class="speaker-text">
- <span v-if="processedTranscription.isJson" v-for="word in segment.words" :key="word.start" class="word" :data-start-time="word.start">${ word.displayWord }</span>
- <span v-else>${ segment.text }</span>
+ <span class="word">${ segment.sentence }</span>
</div>
</div>
</div>
@@ -1000,10 +993,9 @@
<div v-for="(row, index) in processedTranscription.bubbleRows" :key="index"
:class="['bubble-row', { 'speaker-me': row.isMe }]">
<div v-for="(bubble, bubbleIndex) in row.bubbles" :key="bubbleIndex"
- :class="[row.color, 'speaker-bubble', { 'speaker-me': row.isMe }]">
+ :class="[row.color, 'speaker-bubble', { 'speaker-me': row.isMe }]" :data-start-time="bubble.startTime">
<div class="speaker-bubble-content">
- <span v-if="processedTranscription.isJson" v-for="word in bubble.words" :key="word.start" class="word" :data-start-time="word.start">${ word.displayWord }</span>
- <span v-else>${ bubble.text }</span>
+ <span class="word">${ bubble.sentence }</span>
</div>
</div>
</div>