Skip to main content

Your First Transcription

This guide walks you through creating your first transcription in OpenTranscribe, from upload to analysis.

Step 1: Prepare Your Media File​

OpenTranscribe supports a wide range of formats:

Audio Formats​

  • MP3 - Most common audio format
  • WAV - Uncompressed audio (best quality)
  • FLAC - Lossless compression
  • M4A - Apple audio format
  • OGG - Open-source audio format

Video Formats​

  • MP4 - Most common video format
  • MOV - Apple video format
  • AVI - Windows video format
  • MKV - Matroska video container
  • WEBM - Web-optimized video

File Size Limits​

  • Maximum file size: 15 GB (MAX_UPLOAD_BYTES)
  • Recommended: Under 2GB for faster processing
  • Long videos (3+ hours) supported
Best Results

For best transcription quality:

  • Use clear audio with minimal background noise
  • Single speaker per channel if possible
  • Good microphone quality (not phone speaker recordings)
  • Volume normalized - not too quiet or clipping

Step 2: Upload Your File​

Via Web Interface​

  1. Click "Add Media" in the gallery, then stay on the Upload File tab
  2. Drag and drop your file onto the upload zone, or click to browse
  3. Select your file from your computer
  4. Watch the upload progress in the floating upload manager

Drag and drop upload interface

Via URL (YouTube)​

OpenTranscribe can download and process YouTube videos:

  1. Click "Add Media" in the gallery and open the Media URL tab
  2. Paste the YouTube URL (supports playlists too!)
  3. Click "Process Video"
  4. The video will be downloaded and queued for transcription

URL import - paste YouTube or video URL

Via Recording​

Record audio directly in your browser:

  1. Click "Add Media" in the gallery and open the Record Audio tab
  2. Select your microphone device
  3. Click "Start Recording"
  4. Monitor audio levels to ensure good volume
  5. Pause/Resume as needed
  6. Click "Stop" when finished
  7. The recording is automatically uploaded and processed

Upload settings - configure tags, speakers, and model

Step 3: Monitor Processing​

Processing Stages​

  1. Queued - File is waiting in the processing queue
  2. Preparing audio - Extracting the audio track (from video if needed) and converting it for transcription
  3. Transcribing - Turning speech into text, either with the built-in model on this server or with the cloud provider chosen in Settings → Transcription
  4. Detecting speakers - Working out who spoke when, as chosen in Settings → Speaker Identification → Speaker Detection (this server, your transcription provider, the pyannote.ai cloud service, or off). Depending on that choice this happens together with transcription or just after the transcript is saved
  5. Voiceprints and matching - Computing a voiceprint for each speaker on this server and matching it against speakers in your other recordings
  6. Indexing - Adding the transcript to the search index
  7. Complete - Ready to view!

To see, before you upload, where each step will run for you (this server or a named cloud provider) and whether anything blocks it, open Settings → Transcription & Speakers → Processing pipeline.

If an AI provider is configured, summaries, topics and speaker name suggestions follow once the transcript is ready. To see which provider and model handled each step of a finished file, open the file and expand its Processing section. See How Your Files Are Processed.

Where to Watch Progress​

Real-Time Updates:

  • Upload Manager (bottom-right floating panel)
  • Notifications Panel (bell icon in navbar)
  • File Library (processing badge on file cards)
  • Flower Dashboard (http://localhost:5175/flower)

Notifications panel showing transcription progress

How long notifications are kept: the panel is an inbox, so nothing disappears while you are looking at it. Old, successful notifications are cleared when the app loads, when a new one arrives, or just after you close the panel. Errors, warnings, work still in progress, and notices that may need your attention (shares, takedowns, re-indexing, migrations) are never cleared automatically. A successful notification is removed 24 hours after you read it, or after 14 days if you never opened it. At most 100 are kept. Use Keep notifications for at the bottom of the panel to choose 24 hours after reading (default), 7 days after reading, or Never auto-clear; the 100 limit still applies to routine successful notices. If anything was cleared, the panel tells you how many the next time you open it. The setting is stored in your browser.

Processing Time Estimates:

DurationGPU (RTX 3080)CPU (8-core)
5 min~30 seconds~5 minutes
30 min~3 minutes~30 minutes
1 hour~5 minutes~60 minutes
3 hours~15 minutes~3 hours
Processing Speed

With GPU acceleration and the large-v3-turbo model (default), OpenTranscribe processes at ~40x realtime speed (full pipeline). A 1-hour file transcribes in about 5 minutes!

Step 4: View Your Transcript​

Once processing completes, click on the file to view the transcript.

Transcript Features​

File detail page with video player and synchronized transcript

Interactive Transcript:

  • Click any word to jump to that moment in the audio
  • Speaker labels automatically assigned (SPEAKER_00, SPEAKER_01, etc.)
  • Timestamps show when each speaker talks
  • Word-level highlighting follows audio playback

Waveform Player:

  • Click anywhere on the waveform to seek to that time
  • Visual representation of audio amplitude
  • Zoom controls for detailed view
  • Speaker segments color-coded

Playback Controls:

  • Play/Pause audio playback
  • Speed control (0.5x to 2x speed)
  • Volume control
  • Keyboard shortcuts (Space to play/pause, arrow keys to seek)

Step 5: Edit Speaker Names​

OpenTranscribe automatically detects speakers but labels them generically. You can edit names:

Edit Speaker Names​

  1. Click the "Edit Speakers" button below the transcript
  2. Click on a speaker label (e.g., "SPEAKER_00")
  3. Type the actual name (e.g., "John Smith")
  4. Press Enter or click outside to save
  5. The name updates throughout the transcript instantly

Speaker editor with identified speakers and voice match suggestions

Create Speaker Profiles​

When you name a speaker, OpenTranscribe can:

  1. Create a global profile for that speaker
  2. Generate a voice fingerprint using their audio
  3. Suggest that speaker in future transcriptions
  4. Track their appearances across multiple files

See Speaker Management for advanced features.

Step 6: Generate a Summary (Optional)​

If you've configured an LLM provider, you can generate AI summaries:

Configure LLM (One-Time)​

  1. Open Settings from the user menu
  2. Under AI & Chat, choose LLM Provider Configuration
  3. Select a provider (OpenAI, Claude, vLLM, Ollama)
  4. Enter your API key or endpoint
  5. Test the connection
  6. Click "Save"

Generate Summary​

  1. Open a transcription
  2. Click the "Summarize" button at the top
  3. Choose a summary prompt:
    • BLUF (Bottom Line Up Front) - Executive summary format
    • Meeting Notes - Action items and decisions
    • Custom prompts - Create your own!
  4. Watch the progress (takes 10-60 seconds depending on length)
  5. View the summary in the Summary tab

Summary Features​

The default BLUF summary includes:

  • Overview - High-level summary in 2-3 sentences
  • Key Points - Bullet points of main topics discussed
  • Action Items - Tasks and assignments with priorities
  • Decisions Made - Key decisions and outcomes
  • Follow-up Items - Things to revisit or research
  • Speaker Analysis - Who spoke most, key contributions

AI-generated BLUF summary with executive overview and action items

Step 7: Explore Advanced Features​

Search Your Transcript​

Keyword Search:

Search for: "project deadline"

Finds exact matches of that phrase

Semantic Search:

Search for: "budget concerns"

Finds related concepts like "financial constraints", "cost overruns", etc.

Add Comments​

  1. Click anywhere in the transcript
  2. Type your comment in the comment field
  3. Press Enter to save
  4. Comments are timestamped and linked to that moment

Export Options​

Export Formats:

  • TXT - Plain text transcript (configurable: timestamps, speaker names, comments)
  • JSON - Structured data with timestamps and speaker info
  • CSV - Comma-separated values for spreadsheet analysis
  • SRT - SubRip subtitle file for video players
  • VTT - WebVTT subtitle format for web playback
  • Markdown - Notes with YAML front matter for Obsidian and similar tools
  • Word and PDF - Printable documents with numbered lines you can cite by page and line

Export Methods:

  1. Click the "Export" dropdown on the transcript page
  2. Choose your format
  3. The file downloads immediately. Word and PDF first ask for a few options (line numbers, page size) and are built in the background, so the download starts a moment later.

Export format options dropdown

See Uploading Files - Export Options for full details on each format.

Organize with Collections​

Group related files:

  1. Click the "Collections" button
  2. Create a new collection (e.g., "Q1 2024 Meetings")
  3. Add files by clicking the collection tag
  4. Filter your library by collection

Performance Tips​

For Faster Processing​

  1. Enable GPU acceleration if available
  2. Use smaller models for quick drafts (base or medium)
  3. Process overnight for large batches
  4. Multi-GPU scaling for high-throughput needs

For Better Accuracy​

  1. Use large-v3 model for best transcription quality (including non-English and translation tasks)
  2. Good audio quality - clear, well-recorded audio
  3. Edit speaker names to improve future speaker matching
  4. Verify and correct any transcription errors

Troubleshooting​

Upload Fails​

  • Check file size (must be under 15 GB)
  • Check format (must be supported audio/video)
  • Check disk space (need enough storage)
  • Try again - network errors can be temporary

Processing Stuck​

  • Check logs: ./opentranscribe.sh logs celery-worker
  • Check Flower: http://localhost:5175/flower
  • Restart workers: ./opentranscribe.sh restart
  • Check GPU memory: nvidia-smi

Poor Transcription Quality​

  • Improve audio - re-record with better microphone
  • Reduce background noise - use noise cancellation
  • Larger model - switch to large-v3 for better accuracy
  • Language setting - ensure correct language selected

Speakers Not Detected​

  • Check your Speaker detection choice - under Settings → Speaker Identification → Speaker Detection, make sure it is not Off, then look at the file's Processing section to see who detected speakers
  • Check the transcription model - some cloud models return text only or segment timing only; a speaker change inside a segment is then not detected
  • Check HuggingFace token - needed when speakers are detected on this server (the full image), because the speaker models are downloaded from HuggingFace
  • Clear audio - speakers need distinct voices
  • Adjust MIN/MAX speakers - honoured by pyannote.ai, the PyAnnote engine and some cloud providers; this server's default engine counts speakers itself and ignores them
  • Manual editing - edit speaker labels manually if needed

Next Steps​

Now that you've created your first transcription, explore:

Need Help?​

Happy transcribing!