When using Whisper App for multilingual scenarios, you can improve the accuracy with the following options:
- Front Configuration::
- Modify the .env file to add `LANGUAGE_PREFERENCE=zh-CN` (in Chinese, for example) when deploying the project.
- Install FFmpeg to handle audio noise reduction: `brew install ffmpeg` (Mac)/`choco install ffmpeg` (Windows)
- recording technique::
- Maintain a steady distance of 15-30cm to avoid breathing noise interference
- Using lavalier microphone access devices in noisy environments
- Real-time transcription for dialog scenes
- Post-calibration::
- Parameter tuning using the Llama model: `temperature=0.7` balances creativity and accuracy
- For specialized terms, you can add a custom thesaurus file `custom_terms.txt` to the project directory.
- Manually timestamped secondary validation of important segments
Tests show that the Chinese transcription accuracy can be improved from 82% to 93% after using the above method.If you need to process dialects, it is recommended to enable the Whisper-large-v3 model in the Together.ai console.
This answer comes from the articleWhisper App: free speech-to-text & AI note organizer toolThe
May not be reproduced without permission:AI productivity tools " How to optimize Whisper App's transcription accuracy in non-English speaking environments?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil