쉐도잉 연습: What's New in Zephyr 4.5? - 영상으로 영어 말하기 배우기

연습 모드 선택

각 모드는 전체 화면 연습실로 열립니다.

499 문장

문장이 너무 짧거나 길면 Edit를 눌러 조정하세요.

Hey everyone, this is Benjamin from the Zephyr Project, back for a new Zephyr release.

Zephyr 4 .5 is being released today, or at least as you're watching this.

It should be out, and there's lots of new features and things that I want to demo, as I always do.

It's been six months since the last release.

That's our new release cadence, in case you didn't get the memo.

Next one is planned for April of next year, 2027.

This is going to be our next long -term support release, But of course, today we are here to talk about all things 4 .5.

And there's a lot that's happened, probably more than usual.

There's something that the project has been sort of like embracing, I would say, which is AI and AI agents.

This has been just like many other open source projects.

This has had a significant impact on the development.

So we've seen lots of new contributors.

In fact, we have a record trying to look at my notes.

You had nearly 1 ,300 contributors to the release.

Lots and lots of bug fixes, like people using AI agents to help them find bugs.

And of course, new features as well.

And also something interesting, performance improvements.

We've seen many areas and many maintainers, and we will be talking about that throughout the demo.

Something else that's worth highlighting, and let me pull the release notes.

As always, this is going to be where you want to go

if you want to catch up on what's new we also have the migration guide

that accompanies those release notes in case

you come from a previous version of Zephyr you always want to know what API is changed and whatnot

but yeah the release notes I will be walking through them with my own take on them plus some demos

that I've put together like I mentioned but this time I think the fact

that we are having two new processor architectures supported should be probably

it is one of like the top highlights it's not often

that we we get one Zephyr again has been around for many years the the typical processor architectures

that are supported are the super well -known ARM Cortex like

Cortex -M or Intel architectures

Intel x86 32 -bit 64 -bits RISC -V etc etc those kind of like general processor architectures,

but in some verticals, in some industries, there's dedicated silicon, right?

And like dedicated architectures.

One is Infineon Tricore.

If you've been driving a car in the past, sorry, 10, 15 years, it's very,

very, very likely that your engine or your brakes or both are running on the Tricore architecture from Infineon.

So it's, as the name indicates, It's a multi -core architecture, which is all about functional safety and automotive scenarios.

So this is now supported in Zephyr, as in the Zephyr kernel can run on those platforms.

And some of the peripherals of the platform are supported as well.

This is landing in 4 .5 as initial support.

There's still lots of work lined up to make it even better over the next few months.

But you can already try it if you have access and

or as always with Zephyr there's good support in emulators in the form of Kimu so you can run your first

Halo Worlds and whatnot on Kimu.

The Qualcomm hexagon architecture is also supported now so it's a DSP slash NPU,

a neural processor processing unit which typically you will find alongside the Snapdragon SoCs.

And so that's, yeah, for all things, signal processing and hardware acceleration for AI inference and whatnot.

This comes with its own toolchain and compilers, etc. And also supported in Zephyr.

This brings us to, I don't know, like 13 plus, I think, different processor architectures supported in Zephyr.

So pretty unique, pretty interesting.

With that, let's move on to some of the features the also top of the list

because this has been keeping many people busy

and this is also something that's landing in 4 .5

but there's so much lined up

that maybe in 4 .6 there will be another demo

because there's a lot of activity around mpipe or the multimedia pipeline

if you come from a Linux background or

if you're a Linux user

you may have been interacting with a gstreamer the idea is for all things multimedia as the name indicates

you often have many streams video audio

that you need to process right

you need to maybe multiplex them like you need to combine your audio with your video okay fair enough

that's an easy one

but you want to apply some filtering some processing some AI

inference engines to do people detection to do object detection people counting printing.

whatever and you want those pipelines to be efficient to run sometimes in parallel

to not eat up lots of memory especially since i mean it's zephyr

so it's constrained environments so that's what mpipe is all about

and for this one

i have a nice demo lined up showing a pretty complex

pipeline as you can see here on the on the on the diagram

we would want to to analyze different streams in parallel, right?

And so you want to analyze a video coming from, in my case, a video camera,

and you want to run object detection in the form of the YOLO well -known algorithm, and you will detect objects,

people, and based on that you want to overlay some information on top of the video

so that you have like those bounding boxes being drawn and then more, right?

If you detect people, you want to maybe do even more filtering when a person enters the frame, you want to capture the event,

send it to the network and or save it to the disk.

So let me show you actually the code briefly before I show the demo, just to show like just a tiny hint as to how

you would typically get the code in your code editor and how the mPype APIs look like.

Those boxes or like those building blocks that you've seen before

that are all chained to one another input on the left

and output on the right and you feed all your pipeline this way.

This is exactly how the code will look like

when you start assembling your pipeline and have like the T operator to split the pipeline in two,

et cetera, et cetera you end up basically just doing just that.

And you can basically assemble your different building blocks.

And when you have them, you build the wires and you wire everything together.

So that's just like a tiny bit of the code, but I think you get the idea. In fact, let's look at the demo

and let's use it as an opportunity also to mention something that's new in the release and

that I will be using the demo

I will be running on native simulator like I often do for those demos

so the Zephyr running on my Linux PC how do I do video on something

that I mean

if I'm in a simulator

I don't have necessarily a physical webcam attached to a physical board right yet on my PC

I do have a webcam I do have video files

that I could use as an input

so one thing worth highlighting in the context of native sim is

100

that as of this new release many many many more peripherals are supported

101

and you can use them in their emulated form or simulated form

102

and hook them up to your native simulator target

103

so whether it's audio video or anything else that

104

that you see here on the list you can easily add them to your device tree basically and

105

so in my case the video is going to be my

106

webcam is going to be piped into my native sim process

107

so that i can easily visualize it i'm going to use the webcam

108

but you can also pipe a file like a video file

109

which is actually really nice

110

and interesting in the context of continuous integration you want to test

111

that your application behaves like it should when it gets video

112

or audio feed and so you can have repeatable samples

113

and repeatable files that you always feed into your CI and that will or should always give you the same output.

114

So back to the demo, how does it look like?

115

Well, I have again, this is going to run on native sim

116

and so it's going to pick up my video camera and that's what the shell script here does,

117

like the few command lines I need to wire and feed my video cam into the Linux native SIM process.

118

And now this is my demo running.

119

And so this is like that window, that Zephyr display window, that's the same you would get on a physical device, right?

120

If you have an LCD display and the webcam would be whatever camera you have on your actual device.

121

And this is Zephyr.

122

So I have my Zephyr shell here and the application is starting to do apparently people detection or caps or whatever.

123

And I can interact with the shell to do like to interact with the system.

124

My pipeline has several layers, right?

125

And one of them is, well, detecting people through the neural network.

126

Another one is overlaying based on the output on the neural network, overlaying the bounding boxes like that yellow rectangle that you see right here.

127

But there's also more features like maybe applying some special effects.

128

I can tune the image so that it's a plane.

129

Whatever, different effects. And then...

130

There's also the ability to be like, okay, if there is someone in the picture, I want to actually blur them, right?

131

And so that's something that I want to do.

132

It's kind of like a layer that's on or off.

133

And that's exactly what M -Pipe allows you to do.

134

Enabling the privacy mode, I would start basically blurring people being detected in the frame.

135

And this is all happening in real time as you saw or remember maybe in that diagram,

136

we feed the video feed into the pipeline.

137

And first thing we do is we split in two queues.

138

One is the one that's going to be going through the neural network.

139

And this one is a so -called leaky queue, which means that knowing that the neural network will likely take some time to process the images,

140

it's going to do just that.

141

And whenever it's done, it picks up the latest image

142

but it's fine if it's if it loses images you wouldn't want to basically try

143

and process those 30 images per second

144

because i mean the device won't be able to catch up

145

but then there's the other branch or the other leg of the pipeline

146

which is displaying the video in real time and

147

so this one is a queue that's not leaky at all and that's meant to really display everything.

148

And so yeah that's my privacy mode right here and if I start hiding people, let's see if hiding people, putting someone back in the frame,

149

you can see that it's capturing and logging the fact that someone entered the room.

150

That's also something that's part of the pipeline.

151

And so if I quit the application I essentially have on my virtual SD card on my native simulator.

152

I should have saved some files and basically whenever someone enters the picture it's being saved.

153

Let me see I should have right here the yep this is what it captured

154

and so that would end up on the SD card

155

and I could stream to the network etc etc. So I think this is this is pretty impressive.

156

This disclaimer

157

that what you've seen it's using some of the pull requests like I mentioned this is still in development

158

so the the video plugin the display plugin are things for

159

which I mean you might not find the stuff readily available in 4 .5

160

but those are plugins

161

that will be probably merged really soon what you have in

162

4 .5 is everything else like all the plumbing for building your multimedia pipelines

163

so I think this one was pretty interesting right let's go back to the release note

164

and see what we have i mentioned networking let's maybe spend

165

some time on networking the uh the networking subsystem is really rich

166

and has been uh really rich pretty much since since day one

167

but there was a recent presentation at the zafir developer summit

168

where one of the main uh one of the maintainers of the zafir networking subsystem gave an update on everything

169

that's happened over the past few um releases and there's a lot because like I mentioned, thanks to AI, LLMs, etc.,

170

it's easier to implement protocols and make sure that we test them and have the proper test suits in place.

171

So there's many, many new protocols that were added to the release, things like Quick and HTTP 3 support.

172

If you're browsing pretty much any modern website on the internet, browsing the Zafir documentation and whatnot, this is being served over HTTP 3.

173

So you get support for that in Zafir now.

174

It's making more efficient use of the bandwidth of the of your memory etc

175

so that's that's an interesting one ssh you can now ssh into a zafford device

176

or the other way around your zafford device can connect to an ssh server we're going to see

177

that in the demo in in a moment and so on

178

and so forth

179

but before i dive into more of those features one thing

180

that's also highlighted here at the top of the list is something really impressive if you think about it.

181

Just by reviewing the code and using agents to help us review the code,

182

we've been able to identify many bottlenecks in the TCP and UDP implementation as well, to the point where in many scenarios,

183

you will have the low throughput that will be improved by 10 -20 % even on TCP,

184

which is kind of a big deal like it can really make the difference.

185

Also latency -wise, this actually comes with a new benchmark that's available also now for everyone to run,

186

which allows to easily test on Kymu your networking performance.

187

One thing that I did not mention, but it's kind of like, it feels like it's going to be AI no matter what.

188

Yes, we used AI agents to improve the networking stack and more.

189

And in fact, you can even, as of Zephyr 4 .5, have your Zephyr board, your Zephyr application,

190

behave as a Zephyr MCP server.

191

So what is MCP?

192

If you're not familiar, it's the Model Context Protocol.

193

It's a standard for anything, usually an application or some kind of server,

194

to expose and make available so -called tools and services, so that an AI agent,

195

if it feels like those tools might be useful to whatever task it has to do,

196

will call the tools and will use them to do its job.

197

And so what you get in Zephyr 4 .5 is a framework to declare in your system new services,

198

new tools, and probably hook them to some Zephyr subsystems.

199

What I did is basically I've been working on bringing up a new board,

200

which is effectively a flight controller for drones.

201

And so it's packed with sensors.

202

And some of those sensors had the drivers in Zephyr, some not.

203

So I built new drivers for them and I've enabled

204

and built and deployed onto the board an MCP server that's going to be looking at

205

and let's maybe look at the code actually I think

206

it's this guy this yeah this is a board where the MCP server what I want

207

it to do is expose all the all the various sensors

208

that are on the board and basically declare them as tools

209

so that if the AI agent wants to interact with the sensors,

210

it will be able to understand what quantities the sensors manipulate,

211

etc. Well, but what the MCP server does is enumerate all the sensors in the system.

212

And for that, there's actually a new notion in Zephyr 4 .5, which is classes of drivers,

213

like when all the sensor bindings in Zephyr now declare that they are of type sensor,

214

which means that you have macros that allow you to directly know, hey, in my device tree, what are all the sensors that I have?

215

And based on that, well, the MCP server is going to expose that information for the AI agent to consume.

216

and so the demo is

217

that I will be plugging the board to my to my pc over USB one thing

218

that I've done also is that over USB I can have a network interface

219

so in fact the board is going to be visible from my pc over internet

220

I've also enabled just for the sake of showing you that it basically just works.

221

I mentioned the SSH server before.

222

The board that I just connected to the PC

223

exposes its host name over MDNS and so on and so I can easily basically just log into it.

224

But the point is that I'm not sure exactly what sensors I have connected to the board, what they mean, what they do.

225

So this is where I want to use the AI agent for that.

226

So let's try basically to use whatever agent I have, I have Andy to just start asking questions.

227

Let me see.

228

Is this board ready to fly?

229

I mean the control N1 that I've just connected to the PC.

230

Let's see what it has to offer.

231

I might need to speed things up or maybe not.

232

Let's see.

233

It should be able to figure out on its own.

234

No polarizing against the board.

235

I'll read the sensors.

236

And this is where, yeah it starts using the mcp server behind the scenes

237

and it will be telling me i don't know

238

it will help me understand what the board has to offer

239

are these sensors readings accurate are they consistent with what you

240

would expect from a from a magnetometer an accelerometer etc let's see that's an interesting report

241

so no cheating here i actually did this in real time apparently the

242

so there's two imus on the board they seem to agree with one another they seem to sort of confirm

243

that the board is sitting still with like a 9 .80 newton blah blah blah of acceleration etc etc

244

but then apparently the compass is off it's reading 1 .22 goes

245

and it shouldn't i this is actually a new driver

246

so maybe i did something wrong with that driver and i'm guessing

247

that this would certainly mess up my flight control algorithm if this is off

248

So I could, it being an agent, right, and it being able also to access the code of my driver, probably my next question will be,

249

is there a bug in the driver that justifies or explains this?

250

Similarly, the sensor that provides the power supply is supposed to read 5 volts.

251

It doesn't.

252

So there's something wrong here, apparently.

253

And, like, that's...

254

An interesting one, right?

255

Because I have sort of like a generic MCP server deployed on the board.

256

Something that's just here to expose the sensors and make the sensors available as tools.

257

But then the knowledge of what is required for flying a drone, what is good or a bad reading for a compass,

258

etc. This is certainly not something that I would bake in Zephyr.

259

This is something and knowledge that comes from the agent.

260

And yeah, I could sort of like continue on

261

that maybe I will prompt with something else and we will see if that comes up with something post.

262

But can we create a small web -based dashboard, I guess in JavaScript, that would show me the readings of the sensors in real time?

263

And yeah, this one, I will speed things up.

264

I will show you when it's done.

265

I likely will just include the rendering of the app.

266

I've been trying this a couple of times.

267

It usually comes up with a really fancy web -based UI

268

that uses the yeah that pulls the sensors in real time

269

and that helps you understand what your device does

270

and then the sky is the limit I guess like when

271

when you start exposing MCP tools services from your Zephyr board it can be anything

272

that taps into the physical world

273

that taps into the sensors the APIs you can control motors you can control power supplies probably bad things too

274

but it's it's really up to you and

275

so yeah let's see this one I'll keep it on screen for maybe a few more seconds

276

and then you'll see as a small screenshot the final rendering

277

but it usually takes only only a couple minutes and I will just kill it.

278

279

and another one real quick on the networking front this I think

280

this was not showing before I was showing a previous version of the release notes RTP don't miss this one

281

if you do well multimedia as well just like we saw with MediaPipe streaming audio feeds and video feeds over UDP,

282

sending, receiving, we have now support for that in Zephyr.

283

So the really quick demo that I built for this one, it should be fun.

284

This is actually an opportunity to briefly walk you through this list of new driver classes.

285

It's always like, there's always a handful, every Zephyr release.

286

I won't be demoing all of them.

287

I mean, some of them are exactly what it says on the bullet point you have a new API for controlling browsers.

288

So it's kind of like on and off.

289

Maybe it wouldn't make for a great demo anyway, but clock monitor, Dali, Lin, back to, I guess, the automotive topic.

290

Lin is kind of can, I think.

291

It's an automotive serial bus, and so on and so forth.

292

The one that I think is interesting, I mean, they all are, but the one that's going to be demoed together with RTP

293

in just a few seconds is Pulse .io where you have this new API for building

294

and implementing those protocols that are really sensitive to timing.

295

You want to control those fancy RGB LEDs with like really

296

well -timed pulses usually backed with some dedicated IP for that in your silicon.

297

But yeah that's the one I want to use and demo.

298

Let me pull the device on screen maybe first.

299

This is this guy right here.

300

This is an LED matrix.

301

So connected to, you barely see it right here, but I have an ESP32,

302

like a wireless, a Wi -Fi device connected to nothing but Wi -Fi.

303

Like it's not connected to my laptop in any way.

304

It's actually connected to a power bank.

305

And so this is connected to my local network at home.

306

And this is running PulseIO to control the RGB LEDs on that LED matrix.

307

And this is running code where an RTP client is listening

308

to multicast RTP packets coming from somewhere else on the local network and feeding video and audio.

309

Sorry, in this case, audio streams.

310

So what I'm going to do is from my main computer, I will be, let me make sure that you all see the webcam right here.

311

I'm going to be running an app that's going to be streaming

312

or like playing video on my laptop and also pushing the video,

313

sorry, again, pushing the audio feed over UDP.

314

broadcasting or multicasting the audio feed to to the network and so let's see what this does

315

we have main engine start four three two one

316

You get the idea.

317

This works really well.

318

I could have, should have shown multiple devices.

319

I think I also have a mode where it would stream the audio from my, like if I were to just talk into my mic.

320

Yeah, I think that's the one actually.

321

Yeah, there you go.

322

And you can see this might not show up super clearly on the recording,

323

but RTP also gives you the ability to actually synchronize all the feeds.

324

So you can do something that's actually looking really cool.

325

It looks like it's actually synchronized.

326

Maybe I should keep it for the rest of the demo.

327

Yeah, I think this should give you a sense of the breadth of all the things that went into,

328

well, networking in this case, but drivers as well.

329

And there's also a lot about all things beyond all things gateways all things routers um

330

so check it out basically uh moving on like i said

331

i mean no particular order uh let's pick the next one security

332

and sbomb improvements this one i'm going to show a quick demo

333

but first i want to talk briefly through what what happened around the spdx tooling

334

and around maybe also some of the the regulations that

335

that emerged over the past few months maybe we covered it in in 4 .4 i don't quite remember

336

but the cyber resilience act is something that every product makers out there are going to have to be aware of.

337

I mean, you have to be aware of if you're putting a product with so -called digital elements on the EU market.

338

There's lots of things you need to do to carefully manage the provenance of your software, to disclose vulnerabilities when you're being made aware of them.

339

And so how do you do that?

340

Well, you need some tooling.

341

Zephyr has had some tooling around

342

that for many years now in this release

343

we've improved the tooling to to let you do SPDX 3 as bombs

344

so SPDX is an ISO standard

345

which the SPDX 3 version is actually standardized now as well

346

that allows you to capture in a well -described manner the

347

what your software is about what are your dependencies what are the the licenses what are the the CVs

348

and the vulnerabilities that apply to them like using some well -known identifiers for your different software packages.

349

You can match them against public vulnerability databases and so on.

350

So just showing you real quick what that means

351

when you want to generate this S -BOM in your system or for your Zephyr build.

352

You would basically build your application.

353

As usual, the only thing is that you need to enable

354

or to tell the build system that it needs to capture some additional metadata for your build.

355

And once all the metadata is there, you can then ask the west spdx command to generate the output for use.

356

There we go.

357

So we're going to scan all the files for licenses, for dependencies, for build configuration and build options,

358

etc. To the point where you get in your build output folder,

359

you'll get the JSON files that correspond to your SBOM and you can then use some tools,

360

commercial tools, open source tools to load them, visualize them, and analyze them.

361

There is one such tool on the SPDX website.

362

If you go to the visualizer that's being made available there, you can actually visualize your S -BOM.

363

So I'm going to do just that here.

364

I'm going to browse my local file system.

365

You might not see me pick up the files, but trust me those are the same files that I just built.

366

Loading them in the tool and then being able to visualize, analyze the data, visualize the dependencies.

367

We're like what are the sources that I am pulling into my build?

368

I am pulling some libraries from Zephyr, I am pulling some things from the HAL, from the vendor apparently.

369

And I mean you can visualize all those dependencies, licenses associated to them and navigate the whole thing apparently there is this build right here

370

that takes as an input quite a few c files

371

and does what a build step does which is compiling and

372

so you have to get a sense of all the all the options

373

that are there it might i mean some security vulnerabilities might be hidden in there

374

if you didn't tweak your build properly so

375

It's nice to be able to have that and have sort of like this manifest of how you build your application.

376

You get the ability, like all those libraries, modules, packages that you pull into your build.

377

Oftentimes, they will have well -known identifiers, like unique identifiers that.. are the ones that when CVEs and vulnerabilities get reported,

378

those are the ones that would match public databases, right?

379

So the tool, given your SBOM and its contents, can tell you that you might actually be impacted by some vulnerabilities.

380

And so you might want to have a look.

381

I think on this one, I was intentionally using an old version of Zephyr to get some hits there.

382

But yeah, that's basically what you can do with this new SPDX 3 tooling.

383

I highly encourage you to use it.

384

Report bugs as you find limitations in it, just like any other area of Zephyr.

385

Something that's also an addition to Zephyr in this cycle, or at least a rework, is the hardening tool.

386

The fact that when you build an application, there's going to be thousands, I guess kconfig options

387

that you may be able to tweak usually

388

we try to come up in in zephyr with sensible default values for making sure

389

that you get kind of the best compromise in terms of performance security footprint

390

but sometimes as part of prototyping and

391

and bringing up your your application you might end up enabling options

392

that are absolutely not suitable for production like you're exposing some some credentials

393

or you're exposing some information to a potential attacker

394

so the the tool that we've had for uh for a

395

while has been improved so that whenever you have a config option

396

that is not set to what it should it will sort of warn you

397

so um for a given build you can run this command

398

which um i can show you how it looks like on the very same built as before

399

and I can see okay this application I built is this looking like it's providing me

400

and using good values

401

or not apparently not I mean

402

this is kind of this is a debug build out of the box

403

it looks like I should probably disable the the boot banner

404

which usually shows a this is Zephyr 4 .5 welcome to your shell apparently I should remove it

405

because this would give attacker information regarding whatever version I use.

406

Similarly, logging should probably be removed and I mean so on and so forth.

407

You get the idea.

408

If I'm using experimental features it might also not be a good idea

409

or at least I need to be aware that I might be using something that potentially is not rock solid.

410

And also related to security, no demo for that

411

but still worth knowing you want your MCU to be focusing on your actual application

412

and not necessarily running stuff that other peripherals on your system could take care of.

413

Typically crypto, when you do TLS and whatnot, there's usually support in silicon to have hardware acceleration.

414

So in this release cycle, we've introduced hardware acceleration for PSA crypto on many silicon vendors.

415

You have the list here.

416

With that, I think we can probably move on to the next topic or next demo.

417

What else do we have?

418

Simulation, testing.

419

I mentioned those new

420

Native SIM peripherals for audio, video, but there's more, like

421

Wi -Fi, LEDs.

422

You can more easily tap into your existing file system.

423

So check that out kind of related

424

or like the counterpart of what we've seen on native sim

425

but for among others for for keymu targets like

426

if you want your keymu simulator to have access to emulated peripherals over i2c spi

427

or like simulate input drivers block drivers thanks to virtual io

428

this is now possible we still probably need to add some some fancy wrappers

429

and scripts to to to simplify the usage of that

430

but this is whether you're using native sim

431

or keymu basically you now have a lot more freedom

432

and options to route or like to to connect emulated

433

and sometimes real devices to your zafari

434

so that's uh yeah like it's really nice to see how easy

435

and i mean you've seen

436

that in action right like the demo the mpipe demo

437

i was using lots of emulated peripherals i could have done the demo on the board

438

but I was just doing the exact same code on the simulator.

439

Twister also has lots of new improvements to check them out.

440

The networking I think back to networking but this is interesting to see

441

and I would like or hope to see that more in other subsystems.

442

There's usually test suites available for all those protocol stacks out there and for networking

443

There's a test suite called TTCN, which is now...

444

it's available like in the tree there's the ability to run those conformance suites to make sure

445

that basically the zephyr tcp ip and networking stack is follows all the rfcs

446

and all the standards out there and i mean you don't have to trust me

447

or believe me when i say

448

that zephyr does x you can actually run the tests for for yourself

449

which is quite nice if you also a few additions

450

or changes as always we do lots of we eat our own dog food

451

and we have always the we need better tools to do our day -to -day job

452

and they usually end up in the tree and

453

so things like the new the ability to tweak

454

and to enable asserts per per module

455

or per file as opposed to having just this big button

456

and this big red switch for turning them globally on

457

or off this is actually really nice

458

because you can have a better control on on

459

when you you enable them and you can also not over your device

460

and and your your memory by enabling too many assets at the same time

461

so really nice to see lots of new options as well for detecting memory

462

memory faults and memory errors etc

463

and tracing wise also this one totally worth a shout out

464

and you have a nice demo and a nice sample there

465

and in the blog post that goes together with this video you will see

466

that as well but tracing in Zephyr typically would require some dedicated tools

467

or scripts to process the traces depending on what trace format you're using etc

468

but usually the one format that's the sort of the easier to manipulate would be CTF, the common trace format.

469

But historically you would require a Python library to process them that would only work on Linux.

470

With some of the changes in this new version of Zephyr you have a new tool

471

that allows you to visualize the um there's basically a terminal viewer

472

that shows you the traces in real time

473

if you do semi hosting in particular you can uh you can visualize the traces really easily

474

or there's also new tools to do uh per photo export etc etc

475

so i think that's that's really nice um with

476

that let me double check that i've gone through the entire list i think i have

477

please make sure to join us on discord

478

if you're not already on the zaffyr discord

479

this is where you can learn more about what the community

480

is up to there is lots of news being discussed there all of the meetups

481

that are being organized in all the cities worldwide

482

you can also learn about them from our discord channel from our website of course

483

we have a zephyr podcast also which i highly encourage you all to subscribe to

484

and and and listen if you are not already

485

and uh yeah just wrapping things up on the release notes like nearly 200 new boards are supported

486

so it's like it's likely

487

that whatever is the new cool board on the block it's something

488

that you can already just run on zephyr check out like

489

i mentioned also the migration guide like that's an important one we try

490

and make it really complete maybe to the point where a human would be overwhelmed looking at it but

491

if you are starting to use AI agents etc

492

they will love being fed with with this information to help you migrate your existing Zephyr code base

493

so do not hesitate I mean you probably don't even have to say anything

494

they will figure it out by themselves with that like

495

and subscribe the video if you've enjoyed it make sure to to tell me

496

if I forgot anything useful make sure to tell us what is the feature

497

that particularly interests you in this new release

498

and we will be back in six months time for the new Zephyr LTS release.

499

Thank you!

앱 다운로드

유창하게 말하는 데 필요한 모든 것

AI 발음 평가문장마다 점수 확인

IPA 발음 연습모든 소리 마스터

단어 학습나만의 단어장 만들기

단어 게임놀면서 외우기

QR 스캔

Download on theApp Store Get it onGoogle Play

FOR YOU

이 레슨의 어휘와 말하기 포인트

이 말하기 레슨은 영상 “What's New in Zephyr 4.5?”을(를) 바탕으로 합니다. 이 영상에는 섀도잉할 문장 499개와 단어 6223개가 있습니다. 말하는 구간의 길이는 39:10입니다. 화자는 분당 약 159단어로, 일상 대화에 가까운 자연스러운 속도로 말합니다. 단어의 82%가 영어에서 가장 많이 쓰이는 3,000단어에 속합니다. 나머지는 연습 전에 미리 확인해 두세요.

이 영상의 핵심 어휘

영상에 나오는 익혀 둘 만한 단어 15개를 발음, 뜻과 함께 정리했습니다.

단어	발음	뜻
demo 명사	/ˈdɛm.oʊ/	데모
sensor 명사	/ˈsɛn.sɚ/	센서
audio 형용사	/ˈɔ.di.oʊ/	오디오
pipeline 명사	/ˈpaɪpˌlaɪn/	관로, 파이프라인
server 명사	/ˈsɝvɚ/	서버, 봉사기
architecture 명사	/ˈɑː.kɪˌtɛk.t͡ʃə/	건축, 건축학
trace 명사	/tɹeɪs/	흔적, 자국
output 명사	/ˈaʊtpʊt/	출력
protocol 명사	/ˈpɹoʊtəˌkɑl/	의정서
vulnerability 명사	/ˌvʌln(ə)ɹəˈbɪlɪti/	취약성
capture 동사	/ˈkæp.(t)ʃɚ/	포획하다
shell 명사	/ʃɛl/	조가비, 조개
detection 명사	/dɪˈtɛk.ʃən/	탐지, 수색
detect 동사	/dɪˈtɛkt/	찾다, 탐지하다
silicon 명사	/ˈsɪləˌkɑn/	규소, 실리콘

영상에 나오는 구동사

단어	뜻
pick up 동사	줍다
bring up 동사	꺼내다
put together 동사	조립하다

따라 말해 볼 만한 문장

영상에 나오는 짧고 완결된 문장으로, 일상 대화에서 그대로 쓸 수 있습니다:

Let's see what it has to offer.
I will show you when it's done.
I've been trying this a couple of times.
I won't be demoing all of them.
What else do we have?

이 영상의 문법

화자가 가장 많이 쓰는 문형을 영상 속 실제 표현과 함께 정리했습니다.

문형	영상 속 표현
현재완료진행 have/has been + -ing — 전에 시작해 지금도 계속되는 동작	you've been driving · has been keeping · I've been working
현재완료 have/has + 과거분사 — 과거의 일이 지금도 관련이 있을 때	It's been · has been · has had
수동태 be + 과거분사 — 누가 하는지보다 무슨 일이 일어나는지에 초점	being released · is planned · is changed
관계절 who / which + 절 — 사람이나 사물에 대한 추가 정보	say, which is · architecture, which is · queue, which means

주의할 발음

화자는 you're, I'm, I've 같은 축약형과 약화된 형태를 63번 사용합니다. 들리는 대로 짧게 발음하세요.

“sh”와 “zh” 소리: capture /ˈkæp.(t)ʃɚ/, shell /ʃɛl/, detection /dɪˈtɛk.ʃən/, acceleration /əkˌsɛl.əˈɹeɪ.ʃən/, efficient /ɪˈfɪʃənt/
긴 단어 — 강세 위치에 주의: architecture /ˈɑː.kɪˌtɛk.t͡ʃə/, peripheral /pəˈɹɪf(ə)ɹ(ə)l/, simulator /ˈsɪmjəˌleɪtɚ/, vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, acceleration /əkˌsɛl.əˈɹeɪ.ʃən/

한국어 화자가 어려워하는 소리:

/f/ — ㅍ(/p/)으로 바꾸지 말고 윗니를 아랫입술에 대기: peripheral /pəˈɹɪf(ə)ɹ(ə)l/, format /ˈfɔːɹ.mæt/, fancy /ˈfæn.si/, briefly /ˈbɹiːfli/, efficient /ɪˈfɪʃənt/
/v/ — ㅂ(/b/)과 구별하기: server /ˈsɝvɚ/, vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, automotive /ˌɔtəˈmoʊtɪv/, impressive /ɪmˈpɹɛsɪv/, privacy /ˈpɹaɪ.və.si/
/z/ — ㅈ이 아니라 성대를 울리는 /s/: expose /ɪkˈspoʊz/, analyze /ˈæn.əˌlaɪz/, browse /bɹaʊz/, visible /ˈvɪzɪb(ə)l/, presentation /ˌpɹɛzn̩ˈteɪʃn̩/

이 영상으로 연습하는 방법

먼저 말하지 않고 영상을 끝까지 듣고 모르는 단어를 적어 둡니다.
0.75배속으로 한 문장씩 섀도잉을 시작하고, 익숙해지면 보통 속도로 돌아갑니다.
자신의 목소리를 녹음해 원본과 비교하고, demo, sensor, audio 같은 단어에 특히 주의합니다.

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

섀도잉 방법: 단계별 전체 가이드 읽기 →

쉐도잉 연습: What's New in Zephyr 4.5? - 영상으로 영어 말하기 배우기

연습 모드 선택

앱 다운로드

추천 영상

It’s not just you

React Conf 2025 React Native Keynote

Sec Growth DataScience staff meeting Sep 14 2022

Expensive RTX 5090 for LLMs? NO. Use This Instead. (Tesla V100 SXM2 + Z8 G4, #RACERRRZ)

Your AI Agent Starts Every Session Like a New Hire - Killian Carlsen-Phelan, Sonar

十年後的電腦會長什麼樣？黃仁勳：就像今天的手機，最少拿來打電話｜中英字幕

How Anthropic uses Claude in Product Management

Armchair Architects: Resiliency in the Cloud – Core Patterns

SpaceX preparing two Starship flights in parallel - Starbase Summary

Did Google just dethrone OpenAI and Anthropic?

9 Insane Claude Mods I Already Can't live Without

Welcome to the Huberman Lab Podcast