#Transcription Problems

Troubleshoot issues with Whisper transcript accuracy and processing.

#Transcript Is Empty

Transcription completed but no text appears.

#Check Whisper Model Downloaded

1. Go to Settings → Transcription → Model
2. Look for a green checkmark or "Downloaded" label next to your model
3. If not downloaded:
   • Click Download next to the model
   • Wait for download to complete (can take 5-30 minutes depending on model size)
   • Retry transcription

#VAD Sensitivity Too Aggressive

Voice Activity Detection (VAD) may have filtered out all audio as silence:

1. Go to Settings → Recording → VAD Sensitivity
2. Lower the sensitivity (move slider left toward "Low")
3. Re-record test segment or regenerate transcript
4. Check if speech is now captured

Note: Very low voices or soft speakers are more likely to be filtered. Test with normal speaking volume.

#Audio Recording Is Actually Silent

Verify the original audio wasn't captured:

1. Open the meeting
2. Click any transcript segment
3. Press Play to hear original audio
4. If silent, see Recording Issues →

If audio plays but transcript is empty, Whisper model isn't working. See below.

#Whisper Model Needs Reinstall

The model file may be corrupted:

1. Go to Settings → Transcription → Model
2. Click the gear icon next to your model
3. Click Delete
4. Click Download to re-download
5. Retry transcription

#Transcript Is Garbled or Gibberish

Transcription contains nonsensical words, wrong language, or corrupted text.

#Check Audio Quality

1. Play back the recording: Open meeting → click segment → Play
2. Listen for clarity
3. If audio is distorted, crackling, or very quiet:
   • See Recording Issues →
   • Adjust microphone gain
   • Retry in quieter environment

#Try a Larger Whisper Model

Larger models are more accurate and better at noisy audio:

1. Go to Settings → Transcription → Model
2. Choose a larger model:
   • Upgrade: tiny → base → small → medium → large-v3
3. Download the new model
4. Regenerate transcript

Accuracy vs Speed tradeoff:

• tiny / base: Faster but less accurate (better for near-perfect audio)
• medium / large-v3: More accurate (can handle noisy audio) but slower

#Set Language Manually

If Whisper auto-detected wrong language:

1. Go to Settings → Transcription → Language
2. Choose the correct language (e.g. English, Spanish, French)
3. Regenerate transcript

Auto-detection fails sometimes with mixed languages or heavy accents.

#Lower VAD Sensitivity

VAD filtering might be cutting out speech:

1. Go to Settings → Recording → VAD Sensitivity
2. Lower sensitivity to Low
3. Regenerate transcript
4. Check if previously missing words now appear

#Transcript Is Very Slow

Transcription is taking longer than expected (over 2-3 minutes for 30-min meeting).

#Use Smaller Model

Larger models take exponentially longer. Switch to a smaller one:

1. Go to Settings → Transcription → Model
2. Choose smaller model:
   • large-v3 (slowest, best quality) → medium (good balance)
   • medium → small (faster)
   • small → base or tiny (fastest)

Typical times for 30-minute meeting:

• tiny: 5 seconds (CPU), 2 seconds (GPU)
• base: 15 seconds (CPU), 5 seconds (GPU)
• small: 45 seconds (CPU), 12 seconds (GPU)
• medium: 2 minutes (CPU), 20 seconds (GPU)
• large-v3: 5 minutes (CPU), 30-45 seconds (GPU)

#Enable GPU Acceleration

GPU is 5-10x faster than CPU. Check if it's enabled:

1. Go to Settings → Advanced → System Info
2. Look for GPU Status
3. Should show "Metal enabled" (macOS) or "CUDA/Vulkan enabled" (Windows)

If showing "CPU only":

• See GPU Acceleration → for setup

#Enable VAD

VAD filters silence and reduces audio sent to Whisper (~70% reduction):

1. Go to Settings → Recording → VAD
2. Toggle On
3. Set sensitivity to Medium or High
4. Regenerate transcript

With VAD enabled, transcription is 2-3x faster.

#Close Other Applications

Free up system resources:

1. Close browser tabs (Chrome, Firefox use lots of memory)
2. Close large apps (photo/video editors, IDE)
3. Close other music apps
4. Restart computer if system is sluggish

GPU acceleration speeds up if system is less busy.

#Check System Memory

If Clearminutes is running out of memory:

1. Open Task Manager (Windows) or Activity Monitor (macOS)
2. Look for "Clearminutes" or "app_lib" process
3. If "Memory" shows 4+ GB, system is under strain
4. Close other apps or increase available RAM

#Transcript Misses Words

Transcript is incomplete — whole words or phrases are missing.

#Lower VAD Sensitivity

VAD may be filtering speech as silence:

1. Go to Settings → Recording → VAD Sensitivity
2. Lower to Low (more permissive)
3. Regenerate transcript
4. Check if previously missing words return

Trade-off: Lower VAD may include more background noise.

#Try a Larger Whisper Model

Smaller models sometimes miss quiet words. Upgrade model:

1. Go to Settings → Transcription → Model
2. Choose larger model (e.g. medium or large-v3)
3. Download and regenerate transcript

#Check Microphone Levels

Words may be missing because they were recorded too quietly:

1. Go to Settings → Recording → Microphone Gain
2. Increase to 70-80%
3. Re-record test segment
4. Check transcription quality

#Manually Add Missing Words

You can edit the transcript directly:

1. Open meeting
2. Click segment where word is missing
3. Click to edit and add the word
4. Save

Use Cmd+Z (macOS) or Ctrl+Z (Windows) to undo if needed.

#Wrong Language Detected

Transcript is in wrong language or garbled because of language auto-detection.

#Set Language Manually

1. Go to Settings → Transcription → Language
2. Select correct language
3. Regenerate transcript

#For Mixed Language Meetings

If meeting has multiple languages:

• Set the primary language in settings
• Manually edit any segments in other languages
• Consider recording multiple meetings if heavily multilingual

Whisper works best with one primary language.

#Diacritics and Special Characters Missing

Accents, umlauts, or special characters not appearing in transcript.

#This Is a Whisper Limitation

Some non-English characters may be:

• Converted to their base letter (é → e)
• Omitted entirely
• Substituted with similar characters

Workaround:

1. Edit transcript manually to add correct characters
2. Use Find & Replace to fix recurring issues

#Try Large-v3 Model

The latest large-v3 model handles non-English characters better:

1. Go to Settings → Transcription → Model
2. Download and switch to large-v3
3. Regenerate transcript

#Personal Names Not Recognized

People's names are misspelled or not recognized.

#Whisper Limitation

Whisper sometimes misspells proper names and technical terms:

• "TensorFlow" → "tensorflow"
• "API" → "Api"
• Acronyms → expanded or capitalized incorrectly

#Fix Names in Transcript

1. Open meeting
2. Use Find & Replace (Cmd+H / Ctrl+H)
3. Find: "incorrect name", Replace: "correct name"
4. Click Replace All

#Rename Speakers (Pro)

If using diarization:

1. Click speaker name in transcript
2. Type correct name
3. Apply to all segments of that speaker

#Transcription Cuts Off Mid-Meeting

Transcript ends abruptly in the middle of recording.

#Check Recording File

1. Open meeting
2. Click a segment near the end
3. Press Play to verify audio continues
4. If audio cuts off, see Recording Issues →

#Regenerate Transcript

Sometimes transcription fails due to temporary issue:

1. Open the meeting
2. Click Regenerate Transcript
3. Wait for processing to complete
4. Check if full transcript appears

#Check Disk Space

If disk was full during transcription:

1. Go to Settings → Storage
2. Free up space
3. Regenerate transcript

#Too Much Background Noise in Transcript

Transcript picks up background sounds as words (air conditioning, keyboard, etc.).

#Enable Noise Suppression

1. Go to Settings → Recording → Noise Suppression
2. Toggle On
3. Set to Moderate or Strong
4. Re-record test segment

#Raise VAD Sensitivity

More aggressive VAD filtering removes quiet background noise:

1. Go to Settings → Recording → VAD Sensitivity
2. Increase to High
3. Regenerate transcript

#Improve Recording Environment

1. Close windows and doors
2. Turn off HVAC/fans if possible
3. Move away from air vents
4. Use a directional microphone

#Use Large-v3 Model

Larger models are better at distinguishing speech from noise:

1. Go to Settings → Transcription → Model
2. Download large-v3
3. Regenerate transcript

#Regenerate a Transcript

To re-transcribe a meeting with different settings:

1. Open the meeting
2. Click Regenerate Transcript
3. Choose new settings (different model, language, etc.)
4. Wait for processing
5. New transcript replaces old one

Note: Old transcript is saved in edit history (if you regenerated immediately). Undo if needed.

#Check Processing Logs

If you're still having issues, check the logs:

1. Go to Settings → Advanced → View Logs
2. Look for error messages related to Whisper
3. Screenshot errors and email to https://support.clearminutes.app

Include:

• Meeting duration
• Whisper model used
• Language setting
• Error message from logs

#Related

• Recording Quality — VAD and audio preprocessing
• Model Selection — Choosing the right Whisper model
• GPU Acceleration — Speed up transcription
• Recording Issues — Microphone and audio capture problems