Speak Clearly
- Use a moderate, steady pace — not too fast, not too slow
- Enunciate clearly, especially for technical terms
- Avoid mumbling or trailing off at the end of sentences
Reduce Background Noise
- Use the app in a quiet environment
- Consider using a directional microphone or headset
- Close windows and doors if there’s ambient noise
- Mute other audio sources on your computer
Check Your Language Setting
- Go to Settings → Speech Task
- Pick the spoken language explicitly instead of leaving Auto-detect. Auto-detect works, but choosing the language you will speak gives noticeably better accuracy. The app itself notes: “Transcription works best when the spoken language is set correctly.”
- If you switch between languages, update this setting each time
Use Output Style and Dictionary
For specialized vocabulary (medical terms, legal jargon, company names):
- Go to Settings → Dictionary for words, names, acronyms, and recurring corrections
- Go to Settings → Output Style for tone, structure, and short domain instructions
- Keep instructions concrete and avoid long paragraphs
Choose the Right Recognition Profile
| Profile | Accuracy | Speed | RAM Usage |
|---|---|---|---|
| Minimal / Compact | Basic to good | Fastest | ~75-148 MB |
| Everyday / Standard | Good to better | Fast to balanced | ~190-512 MB |
| Extended / Expert / Professional | Highest | Slower | ~1.5-3 GB+ |
Larger profiles can improve accuracy but take more time and memory. Choose based on your hardware and needs.
Use Custom Dictionary
- Go to Settings → Dictionary
- Add words that are frequently misrecognized
- Enable auto-correct to fix common transcription mistakes
Learn more in our dedicated Custom Dictionary guide.
Need to Tune Speed and Quality Together?
If accuracy and speed both feel off, see Adjusting Speed and Transcription Quality — it walks through the three knobs (profile, Quality level, GPU acceleration) in a step-by-step routine.
Still Having Issues?
- Check your microphone settings if audio quality seems poor
- Make sure your system meets the requirements for your chosen model
- [Contact support](mailto:support@weesperneonflow.ai?subject=[Help: ts-003] Low Accuracy) if nothing helps