On the Review & Correct page you can monitor dictation, review dictation transcripts, and listen to audio to troubleshoot issues with dictations and audio equipment. You can also correct dictation transcripts to obtain an accuracy score and help improve dictation accuracy. The Accuracy Optimization Service can perform these tasks for you.
Factors That Impact Dictation Accuracy
While speech recognition technology has made significant advancements, accuracy in dictation remains an inherent challenge due to a vast number of influencing factors. Even the most advanced speech-to-text solutions are subject to occasional misrecognitions due to:
- Speaker Variability: Differences in accents, pronunciation, speech patterns, and pacing can impact recognition accuracy. Even well-trained models may struggle with unique inflections or uncommon speech habits.
- Acoustic and Environmental Factors: Background noise, microphone quality, and positioning all play critical roles in dictation accuracy. Suboptimal conditions can lead to misinterpretations of words and phrases.
- Vocabulary and Context Limitations: Speech engines rely on predefined vocabularies and contextual models. When users dictate uncommon terms, abbreviations, or highly specialized terminology, recognition accuracy may decrease unless custom vocabulary adjustments are made.
- Audio Quality and Transmission Issues: Poor internet connectivity, low microphone sensitivity, or interference can introduce distortions in audio input, affecting the clarity of speech captured by the system.
- Homophones and Similar Sounding Words: Words that sound alike but have different meanings (e.g., "there" vs. "their") rely on contextual understanding, which may not always be perfect.
- Real-time Processing Limitations: While AI-driven transcription engines constantly evolve, real-time processing still faces challenges in understanding nuances, speaker intent, and complex phrasing.
Given these variables, occasional misrecognitions are expected. While singular or one-off errors may not always be traceable or correctable, trends of consistent misrecognition for the same word across multiple dictations should be reported. If users notice a recurring issue with a particular word or phrase, submitting a ticket through the Support Portal is recommended. This allows our team to analyze patterns, update vocabularies, and implement improvements where feasible.
Tools to Support Dictation Accuracy Improvement
Achieving high dictation accuracy requires reviewing audio quality, identifying transcription errors, and making necessary adjustments. nVoq offers two key tools to support this process:
- Review & Correct: This self-service tool allows nVoq Administrator users to listen to dictation recordings, review and correct transcripts, and troubleshoot accuracy issues. Administrators can make manual corrections and improve accuracy scores by identifying common problems—such as poor microphone quality, incorrect volume levels, unclear speech, or missing vocabulary.
- You may find these resources helpful while performing Review & Correct in nVoq Administrator:
- Accuracy Optimization Service: The Accuracy Optimization Service (AOS) automates and streamlines the Review & Correct process by having a team of trained medical transcriptionists review dictations, correct errors, and optimize the user’s profile. This service is beneficial for new users who may struggle with microphone settings, speech patterns, or specific terminology. AOS provides expert recommendations to enhance user-specific accuracy and ensures that corrections are incorporated into the system to improve long-term performance.
Performing Review & Correct
Listen to Audio and Dictation Quality
On the Review & Correct page listen for the following issues in dictation recordings:
- Listen for poor audio quality caused by microphone problems.
- The audio quality on these dictations is “fuzzy” and indistinct. The microphone might be bad, or the speaker may be using the "array" microphone that is built into the computer. The ideal microphone is a noise canceling mic, designed to pick up only the speaker’s voice and filter out environmental sounds.
- Sometimes the system will pick up both the primary mic being used and a secondary mic built into the computer (especially with the laptops or tablet PCs). Having more than one microphone recording the dictations will often cause “fuzzy” and indistinct audio or even a noticeable echo which can degrade accuracy.
- Listen for poor audio quality caused by volume level problems.
- While generally speaking clearly overall, the speaker sounds far away from the microphone which reduces the audio quality. Coach this user to exercise better microphone control and keep a proper distance from it.
- Listen for the user rushing words, not annunciating, or skipping words altogether.
- The speaker may be rushing and running words together. It's possible to get good dictation accuracy while speaking very quickly, but for higher accuracy it is important to maintain separation between words. Coach this user to enunciate words clearly, especially those that sound similar such as "and" versus "an", "discussed" versus "discuss", "depressed" versus "depress".
- Listen for dictations that include names of other clinicians or facility names that may be frequently dictated.
- You can add these to the vocabulary and/or to substitutions, if appropriate.
- Listen for dictations that include clinical terms or drug names that are not returning correctly in the transcript.
- These can be added to vocabulary and/or to substitutions, if appropriate.
- nVoq regularly adds new drug names to the dictionary, but if there are commonly used clinical terms or drug names that are not returning in dictations you can reach out to nVoq Support to have the terms added.
- Listen for dictations that include patient names.
- Coach the user to dictate generically, such as “Patient presents today...”
- For information on how to listen to audio in nVoq Administrator see Detail Review & Correct
Correct Dictation Transcripts to Improve Accuracy & Get Accuracy Score
Applying corrections to dictation transcripts (or rejecting recordings in which a user misspoke or the audio quality was poor) are ways to help users get better results in their dictations.
- Transcripts that are corrected on the Review & Correct page are applied to a user's account sentence modeling in a scheduled job that generally runs overnight.
- Accuracy scores are calculated based on corrections made on the Review & Correct page.
For information on how to correct dictation transcripts in nVoq Administrator, see Detail Review & Correct
Review & Correct API
Please see the following resource for Review & Correct using our API:
The Accuracy Optimization Service
The Accuracy Optimization Service (AOS) allows you to have all these functions performed for you. It's designed to maximize the dictator’s accuracy and user experience with nVoq dictation. It helps new users with adjustment issues specific to them, including microphone issues, dictated names or terms specific to each user, use of commands, unusual speech patterns or pronunciations, etc.
The process starts with a specially trained team of medical transcriptionists who review the user’s dictations over the first few weeks, correct errors generated by the dictation engine, and tune the user’s account to maximize accuracy and usability. Audio quality is assessed and recommendations are made regarding subpar audio which can affect the user’s accuracy and experience, including bad audio equipment or microphones, volume levels, poor microphone control, and background noise. Many vocabulary terms that are not recognized or returned properly by the dictation engine are identified and remedial action is taken to address them. Coaching tips are generated in cases in which the user exhibits techniques that adversely affect accuracy. Finally, all the corrected dictations are added to the user’s profile and used to improve the statistical language model, thus improving the user’s accuracy and performance with dictation.
How to Select the Best Candidates for Accuracy Optimization Service
Users who have completed adequate nVoq Dictation training and are knowledgeable on the dictation platform but are still experiencing dictation accuracy misrecognitions would benefit from user-specific fine-tuning of their dictation profile in the way of Accuracy Optimization Service. To determine if a user would benefit from AOS, follow these steps:
- Perform an Initial Review in Review & Correct
- Listen to 10–20 medically relevant, non-training dictations in Review & Correct. See Listen to Audio & View Transcripts.
- Focus on dictations containing 10–50 words.
- Verify that the audio quality is good, the user speaks at a conversational pace, enunciates clearly, and uses medical terminology.
- Ensure that users selected for the Accuracy Optimization Service adhere to nVoq’s Dictation Best Practices.
By selecting the right candidates for AOS, users can maximize the benefits of this service and improve their dictation performance.
Receiving AOS Feedback & Reviewing Corrections in nVoq Administrator
- Receive AOS Feedback via Email
- Once the one-time paid Accuracy Optimization (AOS) is complete, feedback will be sent via email to the designated Administrative contact.
- The AOS will begin upon submission of the request and will be active for a duration of 30 calendar days from the initiation date of either up to 3,500 words or up to 130 dictations are corrected, whichever comes first.
- AOS support includes adding unrecognized vocabulary terms, setting up substitutions, and providing a final feedback email with coaching tips.
- Users who complete AOS can expect improved recognition accuracy and a better overall dictation experience. This service helps address individual challenges such as microphone issues, user-specific terminology, command usage, and unique speech patterns or pronunciations.
- Review AOS Corrections in nVoq Administrator
- Corrected dictation transcripts and dictation audio can be reviewed in the Detail Review & Correct page of nVoq Administrator.
- Tip: Set the date range to include the dates that the dictations occurred and adjust the Status filter to "Corrected" for easy access.
