Note: Everything said in this paper about the Jabra PanaCast 50 also applies to the Lenovo ThinkSmart Bar 180, as they are essentially the same device with different branding.

Introduction

Microsoft announced Intelligent Speaker at Ignite in March of 2021^[3] and it went into preview in the second half of that year. At initial release, EPOS and Yealink^[4] were the only two manufacturers to produce certified Intelligent Speaker devices.

In the blog announcement from Microsoft, Intelligent Speaker was defined as something that will “…allow attendees to use the transcription to follow along or capture actions, by knowing who in the room said what. Whether you are working remotely or following the meeting in the conference room, you can effectively see who said what during the meeting.” ^[5]

Cool. What does any of that mean?

Why does this thing exist?

How do you set it up?

Any security issues with this?

Wait – I thought this was about identifying the person talking, so shouldn’t it be called “Intelligent Microphone”?

These questions, and many more, will be answered in the following beautifully worded document.

What is Intelligent Speaker?

Intelligent Speaker (now often called speaker recognition) is proprietary Microsoft technology to uniquely identify a person's voice to have accurate speaker attribution in places such as meeting rooms. Put another way, when multiple people are speaking in a conference room, and the transcription feature in Microsoft Teams is enabled, how can the sentences and words of each in-room attendee be attributed to them as opposed to being generically attributed to the meeting space?

Here is a sample transcript of a conversation in a conference room, appropriately named “Epic Conference Room of Awesome”. Note the sentences from this snippet of the transcript are not attributed to a human, but rather to the conference room.

Figure Donuts are delicious

First off, how many people are in this conversation? One person talking to themselves? Two people? Three?

Let's say there are two people in this conversation. Who said what? Am I bringing the donuts^[6] or is the other person bringing them? Who even is the other person?

Wouldn’t it be swell if the transcript showed the name of the person who said each sentence instead of just the room name?

What makes a speaker microphone device an Intelligent speaker?

As of March 2025, just having a microphone connected to Teams Rooms or a BYOD laptop (where the user has a Copilot or Teams Premium license assigned) makes a device an Intelligent Speaker. But this has not always been the case. In the before times, you needed a specific piece of hardware to use the speaker recognition feature.

Which now raises the question: Is a Microsoft certified Intelligent Speaker still needed? Is there any benefit to these certified devices over "non-certified Intelligent Speakers"?

Good question. But first, let's dive into a bit of what makes a certified Intelligent Speaker special.

There are some hints of what makes Microsoft certified Intelligent Speakers so fancy. But not much. It is mentioned that Intelligent Speakers include a 7-microphone array^[7] to help identify the voices of up to ten people in a meeting room. Little more is given regarding hardware requirements. The Jabra PanaCast 50 (P50) has 8 beamforming microphones, so I guess that’s good enough!

Beyond the hardware, there are the Microsoft services on the backend that really provide the magic powers. Microsoft says that they are leveraging the powers of Microsoft Graph to “…provide[s] access to rich people-centric data and insight in the Microsoft Cloud to contextualize the transcription. For example, because we know who the speaker is, the acronyms, names of colleagues, and different words the speaker uses can be more accurately transcribed.”^[8]

Cool.

Not mentioned is all the other magic that needs to happen. For example, it must be able to match up a given voice to you. Or me. There is a cloud-based audio-matching algorithm that must do this. And then a cloud-based Speech-to-text service which converts your spoken words to text so that it can be accurately written to the transcript.

And then there must be a way for us to manually fix mistakes (if we care enough).

As of October 2023^[9], there are now six hardware devices that Microsoft has certified for Intelligent Speaker.

1. EPOS Expand Capture 5
2. Yealink M Speech^[10]
3. Sennheiser TeamConnect Intelligent Speaker
4. Jabra Panacast 50
5. Yealink SmartVision 60
6. Lenovo ThinkSmart Bar 180 ^[11]

Back to the big question: Are the above devices any better? I don't know of any specific testing and results. However, Microsoft has repeatedly said that certified Intelligent Speakers will outperform "non-certified" microphones.  In the official Microsoft documentation, it says:

While we're delighted to extend the capability of speaker recognition to more rooms, it's important to note the quality may not match that of an intelligent speaker certified device. So, it's essential to evaluate the advantages of incorporating an intelligent speaker, especially in crucial spaces where attaining the highest quality transcription and attribution is vital. ¹

It was also mentioned in the announcement of speaker recognition no longer requiring certified hardware. This was written by Christian Schacht - Principal Program Manager at Microsoft who is responsible for the speaker recognition feature.

While we're delighted to extend the capability of speaker recognition to more rooms, it's important to note that the quality may not match that of an intelligent speaker device. Intelligent speakers are designed with multiple microphones to provide high-quality audio, maximizing accuracy in recognition and transcription and boasting an industry-leading reduction of word error rate. In rooms where top-quality transcription and attribution are imperative, it’s worth it to assess the benefits of integrating intelligent speaker hardware certified for Teams.  ²

And then Christian Schacht mentioned it again in a YouTube video explaining this feature and how to set it up. Fast forwarding to the 2:30 portion of this video, Christian says

...so long story short, the certified hardware for intelligent speaker will always be a little bit better because we just get more information [..] out of the device directly. ³

It is ultimately up to you if you want to go the certified hardware route. Just note that for the most accurate transcript, Microsoft believes certified hardware is the best option.

Why is this even a thing?

Why does Intelligent Speaker even exist? Heck – have you ever been on a meeting with transcription enabled?

Until 2024 - I hadn't been. (Thanks Copilot! #SpoilerAlert)

So why? I’ll tell you why: Because it’s cool technology!

OK, that’s not why. The why is boring and I’m trying to pep up this section. And I think I’m failing. And now I’m just wasting your time. So here we go:

Regulated Persons.

There you go. Good times.

What’s a regulated person?

According to some random website called Law Insider, “Regulated Persons means certain broker-dealers and registered investment advisers that are subject to prohibitions against participating in pay-to-play practices and are subject to the SEC’s oversight and, in the case of broker-dealers, the oversight of a registered national securities association, such as FINRA.”^[12]

Put another way, these people have all their communications logged, tracked, and recorded so, should a legal issue arise, they can claim their innocence. Hopefully. Otherwise: jail time.

So how do you track everything someone says when they walk into a common space like a conference room? Hello, Intelligent Speaker.

This has been the primary use case for Intelligent Speaker since its launch. And as such, this has been a niche feature that most Teams admins have either never heard of or have ignored because there was no need for it.

BUT THAT IS ALL CHANGING!

Let me introduce you to my little friend – Copilot! Copilot! Copilot!

What happens if you throw AI, erm, Copilot at a transcript? It can quickly summarize it, pull out notes, and even put together a list of tasks derived from the meeting. And a transcript from a meeting room with proper attribution for Copilot to ingest? That’s like the greatest thing ever.

Now Copilot can summarize tasks like “Michael agreed to buy the donuts” instead of “Conference Room A agreed to buy the donuts”^[13].

The Intelligent Speaker feature now becomes far more than a niche feature for a handful of regulated persons. It becomes a potential game changer for office workers around the globe.

What are the requirements for Intelligent Speaker?

We now know that Intelligent Speaker will save the planet. Or something like that. How does one set it up? What are the requirements? There are quite a few and I’ll start by setting up the administrative side and then show you how to set up the end-user side.

Hardware requirements

I’ve already mentioned the six supported Intelligent Speakers above. But here they are again with pictures. I’m doing this to drive home just how different the Jabra PanaCast 50 (and the Lenovo ThinkSmart Bar 180) is than the other certified Intelligent Speakers on the market.^[14]

​

EPOS Capture 5
Jabra PanaCast 50
Lenovo ThinkSmart Bar 180
Sennheiser TeamConnect Intelligent Speaker
Yealink MSpeech
Yealink MVC S60 (Maybe)

Two are an all-in-one video bar with industry leading video and audio. Three of them are speaker pucks and the other is a center-of-table device. Can you spot which one is the best option?^[15]

Note that Intelligent Speaker now works without the need for a certified Intelligent Speaker. See the discussion above. TLDR: any microphone connected to Teams Rooms or a Copilot-licensed user laptop (PC or Mac) will work, but a certified Intelligent Speaker will work better.

And that’s about it for hardware.

Note: If someone is still using a Logitech SmartDock running Teams Rooms, sell them a more modern Teams Rooms implementation. Remind them that Intelligent Speaker is not supported on those ancient things due to “… a known issue that Teams Rooms can't recognize the Intelligent Speaker through the dock.” ⁴

Software Requirements

The software requirements for Intelligent speaker are straightforward. You will need a Microsoft Teams Rooms (Android or Windows)  installation, connected to the Intelligent Speaker of your choosing Jabra PanaCast 50. You also need to set the PanaCast 50 as the default speaker and microphone within Teams Rooms. For Intelligent Speaker to work, it must be the default speaker and microphone.^[17]

Network Requirements

The network requirements for Intelligent Speaker are the same as for any Teams Rooms installation, with the exception that when using speaker attribution, you need 7Mbs of available upload bandwidth. ^[18] On the nerd side, the seven microphones could send seven streams to Microsoft. That adds a maximum of 1Mbs per audio stream which would be a maximum of 7Mbs. Once the audio streams reach Microsoft, magic happens, and voice matching is tried in the cloud.^[19] 

If you are not using certified intelligent Speaker hardware, then there is no increase in upstream audio bandwidth as only a single stream is sent to Microsoft. It is the same stream of audio sent if you have never enabled this feature. As such, the required audio bandwidth for non-certified Intelligent Speaker microphones is 1Mbps.

Note that the audio required for a Microsoft Teams client joining a meeting with Intelligent Speaker enabled does not change. The same single stream of combined audio is sent to the Teams client if Intelligent Speaker is enabled in the meeting or not.

Licensing Requirements

The Teams Rooms resource account needs a Teams Rooms Pro license assigned to it. Speaker attribution is not supported on the Teams Rooms Basic license.^[20]

Note: If you are still using the legacy Teams Rooms Standard or Teams Rooms Premium license, Intelligent Speaker features will work with both licenses.

For BYOD scenarios, the person connecting the intelligent speaker to their Teams client (aka the "Room Host") needs a Copilot or Teams Premium license. ⁵ The meeting organizer in a BYOD scenario does not need Copilot or Teams Premium, unless they are also the person connecting the intelligent speaker to their laptop (Room Host).

How about user licenses?

The meeting organizer needs at least an E1 or A1 license in order to create the meeting. Their ability to permit recording and transcription is based on the Teams AI policy assigned to them. This gets discussed later.

Attendees whose voice you want attributed in the transcript must be members of the same tenant as the meeting organizer. They also need to be able to sign in to Teams and record their biometric voiceprint. The ability to record a voiceprint is also controller by a Teams AI policy, discussed in the "Create/Edit Teams Policies" section below.

Note that you can't do much with a meeting transcript at this point, at least with regards to Microsoft tools. All you have now is a transcript with speaker attribution. If you want to do something useful with it (at least in this modern Microsoft world), you will need to assign your users Copilot and/or Teams Premium licenses.

Note: You can always feed the transcript into ChatGPT or other large language models and they'll do a really good job analyzing the transcript for you. You don't *have* to use Copilot, it's just easier and native within the Microsoft ecosystem.

For Teams Rooms: The meeting organizer does not need a Copilot or Teams Premium license for voice recognition/intelligent speaker to work.  You can happily join a meeting hosted by someone in your tenant with an E1 license and - if you have a Copilot license - you can then fire off prompts to Copilot during or after the meeting. Your ability to query Copilot is in no way tied to the meeting organizer also having a Copilot license. All Copilot is doing is analyzing the transcript. The meeting organizer's role in your ability to use Copilot only goes so far as their rights to create meetings that allow for transcripts.

For BYOD scenarios, the "Room Host" needs to have a Teams Premium or Copilot license for Intelligent Speaker to work. The room host is the person who connects their laptop to the intelligent speaker (in-room microphone). ⁵

In the scenario above, the meeting organizer (who does not have a Copilot license) cannot use Copilot and will not get an intelligent meeting recap, but you will as you have a Copilot license. I made a table below to help explain this and other licensing add-on scenarios. The below is above and beyond the E1/A1 minimum license.

Note: This is assuming attendees have the right to view and download transcripts. Here is how to block downloads of Teams transcripts.

Note: Downloading transcripts presumes that transcripts are enabled for the meeting. This does not apply to the option to use Copilot without transcripts.

Other stuff

This feature is available in all countries and regions, at least as Microsoft defines them. That does not mean that all languages and locales are supported. See this list for a list of supported locales^[21].

Beyond being available in certain locales, there are legal ramifications to using Intelligent Speaker. For Intelligent Speaker to work, users will have to give up some biometric information (i.e., their voice print). Some nations, principalities, city-states, workers councils, and other political boundaries may have an issue with this. Check first if this is legal to be used where you intend to set it up.

Second, if it is legal to be used, verify with your organization's legal team if it is legal (or desired at all) in your organization. Some companies like plausible deniability. Not having a transcript sure helps avoid some of that pesky legal paperwork that needs to be handed over in a lawsuit. Or they just really value their employee’s privacy.

Assuming the points above are cleared, the Microsoft Teams administrator then needs to create policies that enable the voice attribution feature. Starting in Mid-February 2025, this feature is enabled by default for all users via the global Teams AI policy. Admins can be more tactical by disabling the Global Teams AI policies, creating custom policies, and only assigning the policy to users willing to give up their voice print for the common good.

The ability for users to record and capture their voice and face within Teams is now defaulted to On for every user.  This was announced in Message Center post MC892646.

Which leads to….you must hope your users record their voice prints. It is completely voluntary for them to do this. It’s generally bad form to force an employee to give over personal biometric data like their voice^[22].

How do I set up Intelligent Speaker?

This can get a little tricky but if it were easy, I wouldn’t be writing this.

For only Teams Rooms on Windows, continue below. For Teams Rooms Android and BYOD, you can safely skip down to Create/Edit Teams Meeting Policies.

Configure the hardware

The first thing is to make sure your Intelligent Speaker is on the latest firmware. As I work for Jabra, I'll walk through how to configure our PanaCast 50. If you are using a different vendors certified Intelligent Speaker, follow their instructions for properly configuring the hardware.

Make sure your Jabra PanaCast 50/Lenovo ThinkSmart Bar 180 is on firmware version 6.22 or later. You also need to have Jabra Direct^[23] version 6.11.28601 or later.

Connect your PanaCast 50 to a computer and start Jabra Direct. Once the PanaCast 50 is recognized by Jabra Direct, click on it to get to the settings.

On the screen that appears, click on Settings to get to the good bits.

From within Device settings, scroll down until you see the Playback device type setting. Hit the drop down and change it from “Communication device” to “Microsoft Teams Rooms device”.

Click Save at the top and then reboot. The PanaCast 50 is now ready to be an Intelligent Speaker.

Configure Teams Rooms on Windows

After your Jabra PanaCast 50 reboots, you’re not quite done. You now need to verify that the setting was successfully applied and that it is set as the correct output device within Teams Rooms on Windows.

Go to the Teams Rooms on Windows console and tap More.

On the next screen tap Settings.

You are then prompted to sign into Teams Rooms with administrative credentials. Enter the administrative password to move on to the next step.

From within Settings, scroll down to the Peripherals section.

Finally, set the Audio settings for Teams Rooms. For Microphone for Conferencing, select the PanaCast 50 that has UAC2_TEAMS in the name, as shown in the image below. (For a ThinkSmart Bar 180, the name will be different, but (UAC2_TEAMS) will be the same).

Set the Speaker for Conferencing to the PanaCast 50 that has UAC2_Render in the name. This is shown in the below image. (For a ThinkSmart Bar 180, the name will be different, but (UAC2_Render) will be the same).

Set the default speaker to the same thing you set above – the (UAC2_Render) device.

At this point you’ve completed the easy part from the admin side. Now you need to create some policies.

Create/Edit Teams Policies

Up until now, this has been straightforward and anyone with a laptop, a cable, and (optionally) a PanaCast 50 handy can do this work. At this point, things change. In most organizations, you now need to bring in your Microsoft 365 administrators as you need to edit or create new policies to apply some custom settings.

Microsoft has a PowerShell cmdlet called CsTeamsAIPolicy. This cmdlet provides granular control over voice and face recognition and sets the stage for future AI-related settings.

Within CSTeamsAIPolicy, the Global policy is set with the parameter EnrollVoice (and EnrollFace) set to Enabled.

The person creating or editing these policies needs any of the following roles assigned to them:

• Teams Administrator^[24]
• Teams Communications Administrator^[25]

The Global Teams AI policy only enables voice (and face) enrollment of your end users. You still need to create a new policy for the Microsoft Teams Rooms Resource Account and a policy that is applied to your users allowing them to enable transcription.^[26]

The first policy I will create is to enable the speaker attribute feature. For Teams Rooms, note that you don’t set policies on the Teams Rooms device, you set policies on the Resource Account that signs into Teams Rooms and runs the meetings on the device.

First, I will create a policy called IntelligentSpeakerMTR that sets the value “roomAttributeUserOverride” to “Attribute”.^[27]

There are three values you can set for “roomAttributeUserOverride”.

One is “False” which turns the feature off. Another is “Attribute” which enables speaker attribution. The third is “Distinguish” which tells the speaker to distinguish between different voices but to *not* provide name attributes for the transcript (e.g., “Speaker 1”, “Speaker 2” instead of “Alice”, “Bob”)

Create Policies

Creating the End User Policy

To enable or disable a users' ability to register their voice (and face), you will work with the  CsTeamsAIPolicy.

Note - Be sure to routinely update your Teams PowerShell modules.

Update-Module -Name MicrosoftTeams

If you run Get-CsTeamsAIPolicy, it should return a single, global policy where both voice and face recognition are enabled.

If that is what you want in your organization - Great! You are done. Scroll down to the next section (Assigning Teams Rooms Policy).

To allow flexibility in your environment, create two additional policies: one that enables only voice enrollment (disabling face enrollment), and another that enables only face enrollment (disabling voice enrollment).

To do this, run the following two cmdlets:

New-CsTeamsAIPolicy -Identity VoiceOnly -EnrollFace Disabled -EnrollVoice Enabled

New-CsTeamsAIPolicy -Identity FaceOnly -EnrollFace Enabled -EnrollVoice Disabled

New-CsTeamsAIPolicy -Identity EnableAll -EnrollFace Enabled -EnrollVoice Enabled

If you're wondering why that last one is created when it is a duplicate of the Global policy, keep reading. 🙂

Within policies in Teams, there is the concept of a policy hierarchy. When there are conflicting policies for a user, the policy *closest* to the user is applied.

In the above, there is a global policy that enables both voice and face enrollment. But we just created a policy enabling only one or the other. Which policy takes precedence? The easiest way to think of this is that since you will specifically apply one of the two policies we created to a user, those take precedence over the Gobal policy. For the third one to make sense, change the Global policy values from enabled to Disabled.

Set-CsTeamsAIPolicy -Identity Global -EnrollFace Disabled -EnrollVoice Disabled

I'm personally opposed to ever editing the Global policies. The upside to editing global policies is it's easy. Change it there, and everyone gets the policy. The downside is you lose some flexibility. It's up to you and your IT policies in how to handle editing global policies.

If you choose to leave the global policy untouched, you need to create one more policy to disable both voice and face enrollment.

New-CsTeamsAIPolicy -Identity DisableAll -EnrollFace Disabled -EnrollVoice Disabled

Now assign that policy to all your users. Then for the users who are authorized to enroll their voice or face, assign one of the three policies above.

To remove a policy, use the Remove-CsTeamsAIPolicy. The following cmdlet removes the EnableAll policy.

Remove-CsTeamsAIPolicy -Identity EnableAll