Comparison of AI-Based Transcription Services for Healthcare Professionals

The article is updated on an ongoing basis to ensure that the information is relevant and accurate. The vendors mentioned have been informed of the comparison and have been given the opportunity to contribute updated information.

1. Functionality

AI-based transcription services for healthcare professionals primarily serve two purposes:

1. Automated transcription: The system listens to the conversation between doctor and patient and converts the speech into text in real time or after the consultation.
2. Generation of structured medical record notes: AI helps to reformulate the transcription into a more readable and medically accurate record text.

Comparison of features

Feature	MediVox	Noteless	Medbric
Transcription of doctor-patient conversations	✅	✅	✅
Automatic medical record note generation	✅	✅	✅
Speaker separation (doctor vs. patient)	✅	❓	❓
Meeting minutes	✅	✅Dictation	✅
Personalisation of medical record notes	✅	✅Phrases	❓
Use of generative AI (LLM)	Open AI (with ZDR), Mistral and others	❓	Not specified in its data processing agreement

Note: Not all information about specific features is publicly available for every solution. We recommend that interested users contact the vendors directly for a detailed overview. Technical information has either been submitted to Nettsak by the vendors, or obtained from publicly available sources. For MedBric, information has been obtained from their publicly available data processing agreement.

Pricing structure

Costs and commitment periods are important factors for many clinics. Prices may vary depending on usage patterns and contract terms. The price comparison is based on unlimited use with the various players, as stated in the available information.

Solution	Price per month (with commitment)	Price per month (without commitment)	Commitment period
MediVox	299,-	349,-	12 months (optional)
Noteless	1,200–2,000,-	❓	12 months
Medbric (external use)	-	1,499,-	No commitment
Medbric (EG InfoDoc)	-	2,090,-	No commitment

Sources: Prices for MediVox and Medbric have been obtained from their official websites. The prices for Noteless are based on tips from users and have not been verified by Nettsak. EG InfoDoc sells MedBric integrated into InfoDoc and has published its prices here. We recommend contacting Noteless directly for official prices.

Both MediVox and MedBric have several pricing packages. MediVox has an entry-level package starting at NOK 600 per month (Light - unlimited use. Open for pre-signup) and MedBric has an entry-level package limited to 20 consultations per month, starting at NOK 499 per month (Light AI Assistance).

---

2. Security and privacy

Key questions about security

To ensure the safe use of AI tools in healthcare, it is important to consider the following:

• Storage: Does the solution store transcriptions or medical record notes?
• Technology: Which transcription technology is used?
• Deletion: How long is the data stored before it is deleted?
• Server location: Where are the servers located (Norway, the EU or a third country)?
• Security measures: What measures have been implemented to protect patient data?

Secure handling of patient data is crucial when using AI in healthcare. According to Section 3-1 of the Patients' Rights Act, patients have the right to information about how their data is processed. This means, among other things, that patients should be informed about the use of AI tools during consultations.

Transcription methods and security

There are several technologies for speech-to-text transcription, which vary in accuracy, security and areas of use:

• Azure Cognitive Speech: Microsoft's cloud-based solution that uses advanced neural networks for high precision. It supports multiple languages, has built-in noise reduction and can be fine-tuned with domain-specific models. Requires a connection to Azure servers, which may have privacy implications.
• AWS Transcribe: Amazon's speech recognition service, with the option of custom vocabularies and medical transcription. Data is processed in the AWS cloud and can be encrypted, but is stored temporarily for quality improvement.
• Google Cloud Speech-to-Text: Uses machine learning for real-time and batch transcription. Supports multiple languages and domains, but shares challenges related to data security when processing in the cloud.
• Whisper V3 (OpenAI): An open-source model that offers high accuracy and can run locally or in the cloud. Running locally provides increased security, while cloud-based solutions introduce privacy risks.

Live and semi-live transcription are often regarded as safer options as they limit the amount of data stored at any one time. The Norwegian Directorate of Health recommends methods that ensure minimal data storage and compliance with privacy rules.

Comparison of vendors' transcription technology

Vendor	Live transcription	Semi transcription	On-Prem transcription
MediVox	Azure Cognitive Speech (Norway East)	Whisper V3 (Norwegian data centre)	Yes
Noteless	❓	❓	❓
MedBric	Amazon Web Services (AWS): Provides language and speech-to-text models. EU/EEA Google Cloud Platform: Provides language and speech-to-text models. EU/EEA Microsoft Azure: Provides language and speech-to-text models. EU/EEA	❓	❓

MediVox has specified its technology publicly. MedBric uses several cloud-based vendors for language and speech recognition, but has not specified details about semi- or On-Prem transcription. Source: Data processing agreement publicly available from MedBric, listing sub-processors. Noteless has not responded to the inquiry and has no publicly available data processing agreement.

Anonymisation vs. Pseudonymisation

• Anonymisation: The process where identifiable information is removed completely and cannot be linked back to the individual. Data cannot be reconstructed or traced.
• Pseudonymisation: Identifiable information is replaced with codes or pseudonyms, but can be recreated using a key. This means that medical record notes can recover personal details after transcription.

Noteless states that it anonymises data, which means that the information cannot be traced back to the patient. MediVox pseudonymises data, so that the information can be re-identified when needed. MedBric has not answered how it handles this.

Security when sending data to large language models (LLMs)

When transcribed medical data is sent to a large language model (LLM) for record generation, several security challenges arise:

• Data control: If data is sent to external servers, there may be a risk of unauthorised access or storage.
• Use for training: Some vendors use customer data to improve their models, which may breach privacy rules.
• Guidelines and compliance: The GDPR requires explicit consent and the safeguarding of personal data when processing in AI models.

Overview of vendors' practices:

Vendor	Data protection	Use of LLM	Data storage	CE approved
MediVox	Pseudonymisation	Has a ZDR agreement with Open Info and Mistral	Does not store data	CE Approved
Noteless	Anonymisation	Not answered	Stores data. See noteless.no	Not specified
MedBric	Does not specify anonymisation or pseudonymisation.	Does not specify use of LLM.	Stores data. See data processing agreement. EU/EEA	Not specified

To ensure the safe use of AI in the healthcare sector, it is crucial to choose solutions that follow recommended security protocols and comply with regulatory requirements. MedBric does not specify the use of anonymisation or pseudonymisation, but specifies other security measures in its data processing agreement.

Data storage refers to the storage of the transcription and any generated medical record note. Storing data increases usability but also increases risk. CE approval is addressed in the Norwegian Directorate of Health's guide. No information is currently available about any of the vendors. The case is being updated.

---

3. Responsibility when using AI in healthcare

According to Section 4 of the Health Personnel Act, healthcare professionals are responsible for the proper exercise of their profession. This means that AI solutions used in healthcare must meet strict requirements for security, accuracy and privacy. The Norwegian Directorate of Health emphasises the following key principles:

Quality assurance of AI-generated transcriptions and medical record notes

• Medical record notes and transcriptions generated by AI must be reviewed and validated by healthcare professionals before they are used.
• Any errors or misunderstandings must be detected before the information is entered into the patient record.

The user's responsibility for choosing a solution

• The Norwegian Directorate of Health only provides guidance, and it is the user who is responsible for choosing the right solution for their needs.
• Healthcare professionals must understand the solution's benefits, limitations and risks before implementation.

Prevention of medical errors

• AI-generated texts must always be regarded as assistive tools and not a replacement for professional judgement.
• Errors in transcriptions or medical record notes can have serious consequences and must be detected before further use.

Why security is crucial

Accuracy and patient safety

• Incorrect medical record notes can lead to misunderstandings, mistreatment or medication errors.
• The Norwegian Directorate of Health recommends implementing quality assurance procedures for AI-generated texts.

Privacy and data protection

• Health information is sensitive and is covered by strict rules in the GDPR and the Health Personnel Act.
• Solutions that use AI must have robust mechanisms for privacy and data handling.

AI hallucinations and misinformation

• Studies show that some AI models can generate information that was not said or misinterpret the context.
• This underlines the importance of combining automated transcription with human review.

Compliance with regulatory requirements

• The use of AI in the healthcare sector must comply with the Norwegian Directorate of Health's recommendations.
• Technology vendors must document how their solutions meet legal and ethical requirements.

Open Info's initiative

Open Info is working on a solution that helps customers from all vendors create compliance documents. This is intended to provide a better understanding of the opportunities and risks associated with the use of AI in healthcare, while ensuring that solutions are in line with regulatory requirements.

For more information about quality assurance of AI in healthcare, see the Norwegian Directorate of Health's report.

Accuracy: Errors in transcription and medical record notes can lead to misunderstandings and mistreatment.
Privacy: Health information is sensitive and must be processed in accordance with the GDPR and the Health Personnel Act.
AI hallucinations: Studies show that AI tools can generate information that was not said, which underlines the need for quality assurance.

More information: The Norwegian Directorate of Health's report on quality assurance of AI in the health and care service.

---

5. Conclusion

AI-based transcription tools have the potential to reduce administrative work in the healthcare sector, improve documentation and streamline the workflow. Nevertheless, it is crucial that healthcare professionals consider security aspects, privacy risks and regulatory requirements before choosing a solution.

📌 Healthcare professionals should:

✔️ Carefully evaluate functionality, security and compliance with regulatory requirements.
✔️ Ensure that the solution complies with the GDPR, the Health Personnel Act and the Norwegian Directorate of Health's recommendations.
✔️ Always quality-check AI-generated notes before record-keeping to avoid mistreatment.
✔️ Be aware of possible risks, including transcription errors and AI hallucinations.

Open Info is working to help users from all vendors understand the risks and opportunities of AI in healthcare, as well as to create compliance documentation that simplifies compliance with laws and regulations. If you have questions or need guidance, Open Info can be contacted for advice.

📊 Your feedback is important

To gain better insight into how healthcare professionals experience these solutions, we are conducting an anonymous survey. The purpose is to map which aspects of AI-based transcription tools work well, which challenges exist, and which risks should be addressed.

🔹 The survey is anonymous and is used only for general analysis.
🔹 The purpose is not to evaluate or criticise individual vendors, but to create an overall picture of usage.

We would greatly appreciate it if you could take a couple of minutes to fill out the questionnaire below. Your opinion contributes to safer and better solutions for everyone!

---

About this comparison

This comparison has been prepared to give healthcare professionals an overview of various AI-based transcription services in line with the Norwegian Directorate of Health's guide for the safe use of AI in the healthcare sector (January 2025). The aim is to provide an objective and up-to-date basis for evaluating these solutions.

Collection of information

• Technical information has been obtained from the vendors' official websites and their publicly available data processing agreements.
• All vendors were contacted by email and invited to contribute additional information. The items in the table marked with a question mark (❓) are because we have not received an answer from the vendor, not that the feature is necessarily missing.
• The article is open for updates. If the other vendors wish to contribute more information, the table will be updated to provide as complete a picture as possible.

The author's role

This article was originally written by an external author, but had to be updated due to errors in the first version. I, Håkon Berntsen, was therefore asked to review and correct the content to ensure professional accuracy. I am a systems architect at MediVox.ai and have a role on the company's board. To ensure that the comparison is factual and correct, all information has been obtained from official sources and is not based on subjective assessments.

Do you have updates or input? We encourage all vendors to contribute any corrections or additions, so that healthcare professionals can have the most complete and up-to-date overview. Please get in touch if you have information that should be included.

Sources and references:

• The Health Personnel Act (Section 4) – lovdata.no
• The Patients' Rights Act (Section 3-1) – lovdata.no
• The GDPR and the data protection regulation – datatilsynet.no
• Reports on AI in healthcare – helsedirektoratet.no
• Prices for MediVox – medivox.ai
• Prices for Medbric – https://medbric.com/signup
• Data processing agreement for MedBric - https://medbric.com/databehandleravtale
• Prices for Noteless – Based on submitted tips (not verified by Nettsak)
• The Norwegian Directorate of Health's guide - https://www.helsedirektoratet.no/rapporter/rapport-om-kvalitetssikring-bruk-av-kunstig-intelligens-i-helse-og-omsorgstjenesten