Table 2.
Examples of speech recognition technology in the application of electronic medical documentation.
| Institute | Application scenario | Technical description | Application effect | Ref. |
|---|---|---|---|---|
| Zhejiang Provincial People's Hospital | Generate and extract pathological examination reports: 52h labeled pathological report recordings. | ASR system with Adaptive technology | Recognition rate = 77.87%; reduces labor costs; improves work efficiency and service quality | [81] |
| Western Paraná State University | Audios collected from 30 volunteers | Google API and Microsoft API integrated with the web | Reduces the time to elaborate reports in the radiology | [89] |
| University Hospital Mannheim | Lab test: 22 volunteers; Filed test: 2 male emergency physicians | IBM's Via-Voice Millennium Edition version 7.0 | The overall recognition rate is about 85%. About 75% in emergency medical missions | [77] |
| Kerman University of Medical Sciences | Notes of hospitalized Patients from 2 groups of 35 nurses | Offline SR (Nevisa) Online SR (Speechtexter) | Users' technological literacy; Possibility of error report: handwritten < offline SR < online SR | [74] |
| University of North Carolina School of Medicine | 6 radiologists dictated using speech-recognition software | PowerScribe 360 v4.0-SP2 reporting software | Near-significant increase in the rate of dictation errors; most errors are minor single incorrect words. | [79] |
| King Saud University | CENSREC-1 database: 422 utterances spoken by 110 speakers | Interlaced derivative pattern | 99.78% and 97.30% accuracies using speeches recorded by microphone and smartphone | [18] |
| KPR Institute of Engineering and Technology | 6660 medical speech transcription audio files and 1440 audio files from the RAVDESS dataset | Hybrid Speech Enhancement Algorithm | Minimum word error rates of 9.5% for medical speech and 7.6% for RAVDESS speech | [80] |
| Simon Fraser University |
Co-occurrence statistics for 2700 anonymized magnetic resonance imaging reports | Dragon Naturally Speaking speech-recognition system; Bayes' theorem | Error detection rate as high as 96% in some cases | [83] |
| Graz University of Technology | 239 clinical reports | Semantic and phonetic automatic reconstruction | Relative word error rate reduction of 7.74% | [25] |
| Zhejiang University | Radiology Information System Records | Synthetic method | About 3% superior to the traditional MAP + MLLR | [49] |
| Brigham and Women's Hospital | Records of 10 physicians who had used SR for at least 6 months | Morae usability software | Dictated notes have higher mean quality considering uncorrected errors and document time. | [75] |