header logo icon

Download "Transcribe Text via SFSpeechRecognizer (Lesson 05)"

input logo icon
Go to download
Download this video with UDL Client
  • Video mp4 HD+ with sound
  • Mp3 in the best quality
  • Any size files
"videoThumbnail Transcribe Text via SFSpeechRecognizer (Lesson 05)
play-icon
Similar videos from our catalog
|

Similar videos from our catalog

Video tags
|

Video tags

swift
swiftui
combine
xcode
ios
app
development
programming
code
apple
academy
tutorial
build
iphone
Subtitles
|

Subtitles

subtitles menu arrow
    Download
    00:00:00
    In today's lesson, we will be using SF
    00:00:02
    Speech Analyzer, a new API to allow
    00:00:06
    voice input for our local AI chat app.
    00:00:10
    This topic is a bit more complex to
    00:00:13
    implement as we won't really be living
    00:00:14
    in Swift UI land for this lesson. So,
    00:00:18
    I've already prepared all the code and I
    00:00:19
    will walk you through it during the
    00:00:21
    lesson. Parts of the code are also
    00:00:23
    linked in the video description for your
    00:00:25
    own reference as I won't go through
    00:00:28
    every single detail today. So basically
    00:00:30
    there are three things we need to do.
    00:00:32
    First of all, we of course need to
    00:00:33
    implement a transcription service and
    00:00:36
    I've called it speechtoext service in
    00:00:38
    this project. Then we need to add this
    00:00:41
    little dictation button down here. So
    00:00:43
    this is just an SF symbol button. And
    00:00:46
    then lastly, we need to pull everything
    00:00:48
    together and use the speech analyzer in
    00:00:52
    our existing business logic. So let's
    00:00:55
    get right started. In our UI, we already
    00:00:58
    have our safe area insert for the bottom
    00:01:01
    here with our text field and our send
    00:01:04
    button. So these two things down here.
    00:01:07
    And then we also added a second button
    00:01:11
    with a microphone. So while we are
    00:01:14
    recording and this just basically flips
    00:01:16
    a a boolean is recording, it shows one
    00:01:18
    SF symbol and then let me quickly turn
    00:01:22
    that on. So this is the SF symbol while
    00:01:24
    we are recording and then this is the SF
    00:01:26
    symbol when we stop recording. Of course
    00:01:28
    microphone stuff doesn't work in the
    00:01:30
    simulator. So you'll have to run that on
    00:01:32
    your real device. And before I forget
    00:01:34
    about it, we'll of course need to have
    00:01:36
    this privacy microphone usage
    00:01:38
    description string in our info.plist
    00:01:40
    file in order to get the hardware access
    00:01:42
    needed for this feature. So now that we
    00:01:44
    have our little microphone button in
    00:01:46
    there, we either call start recording or
    00:01:48
    stop recording depending on what the
    00:01:50
    current state is. And I've moved all of
    00:01:53
    the new variables and both of the
    00:01:56
    functions to the bottom of the file just
    00:01:58
    for our reference. But I would advise
    00:02:00
    you to keep all of the state variable
    00:02:02
    and also the speech to text service at
    00:02:04
    the top of the file. So you have a nice
    00:02:06
    structure. So basically we have our is
    00:02:08
    recording flag. Then we have a potential
    00:02:11
    error message to show. We also have a
    00:02:14
    transcription task and we need to store
    00:02:16
    this here but we'll get into that in
    00:02:18
    just a second. And we also have our
    00:02:20
    speechtoext service which we'll also go
    00:02:23
    through in a minute. So then we have our
    00:02:25
    start recording function and our stop
    00:02:27
    recording function. Let's start by
    00:02:29
    looking at the start recording function.
    00:02:31
    What does it actually do? Because we're
    00:02:33
    actually streaming audio here. So we
    00:02:35
    make sure that we're not recording
    00:02:36
    already. So this button can't be pressed
    00:02:38
    twice. This is already somehow checked
    00:02:41
    by the UI or by the button itself, but
    00:02:44
    just to make sure nothing bad happens
    00:02:46
    here. And then of course we set is
    00:02:48
    recording to true. We create our
    00:02:50
    transcription task and this is just a
    00:02:52
    swift concurrency task but we're storing
    00:02:54
    it here in the variable so it doesn't
    00:02:56
    get cancelled when this function is over
    00:02:59
    because we are streaming audio. So this
    00:03:01
    task uh lasts for quite a long time
    00:03:04
    potentially until we cancel it manually
    00:03:06
    in our stop recording function. So in
    00:03:08
    there we await authorization to use the
    00:03:11
    microphone and speech to text. Then we
    00:03:15
    create a stream and we'll look into what
    00:03:17
    this transcribe function does in a
    00:03:19
    minute. We have a four try await. So we
    00:03:22
    are streaming in the partial results
    00:03:25
    from the microphone and we're assigning
    00:03:28
    self.input to this new value and
    00:03:30
    self.input is just the variable that our
    00:03:33
    text field writes to and the variable of
    00:03:36
    the message that we send to our large
    00:03:39
    language model. So basically we take the
    00:03:41
    partial result and we assign it to our
    00:03:43
    input. That way it also automatically
    00:03:46
    gets displayed in the text field and if
    00:03:49
    there's any error we will also assign it
    00:03:51
    to our error message and you can show
    00:03:53
    that on your screen whichever way you'd
    00:03:55
    like. Since we're already here let's
    00:03:56
    also briefly look into the stop
    00:03:58
    recording function before we go into how
    00:04:00
    the transcription using as a speech
    00:04:02
    analyzer actually works. So we make sure
    00:04:05
    that we are actually recording then we
    00:04:07
    set it to false. We cancel and nil out
    00:04:10
    our transcription task. So making sure
    00:04:13
    that no new audio gets streamed into our
    00:04:15
    text field variable. And then we also
    00:04:18
    call stop transcribing on our speechto
    00:04:20
    text service to clean up everything over
    00:04:22
    there. So let's have a look at our
    00:04:24
    speechtoext service. And this file is
    00:04:26
    linked via a GitHub gist in the video
    00:04:29
    description. So you can check it out and
    00:04:31
    download it over there or reference it
    00:04:32
    for your own implementation. At the top
    00:04:35
    here we have two rappers for some old um
    00:04:38
    completion handler based APIs for SF
    00:04:40
    speech recognizer and AV audio session.
    00:04:43
    I won't go into detail. You can look
    00:04:44
    these up in the GS just these are just
    00:04:46
    concurrency wrappers to make our API a
    00:04:49
    bit nicer to use. As I said, I won't go
    00:04:52
    through this line by line because it is
    00:04:54
    quite complex, but I will try to tell
    00:04:56
    you the logic behind how it works. And
    00:04:58
    then if you want to go into it yourself,
    00:05:00
    you're free to check out the file or the
    00:05:02
    documentation that I linked here on top
    00:05:04
    of the file. So, as you can see, there
    00:05:06
    is quite a bit of code in here. So, we
    00:05:08
    have an authorize function. We have a
    00:05:11
    transcribe function. This is probably
    00:05:13
    the most uh interesting one. We have a
    00:05:16
    stop transcribing function, reset, and
    00:05:19
    prepare. engine. So let's look into
    00:05:22
    transcribe because this is the only
    00:05:24
    function that actually gets called in
    00:05:26
    our UI to do the transcription. So if
    00:05:28
    you remember from our UI implementation
    00:05:32
    here we have a stream that we get from
    00:05:34
    the speechto text service transcribe
    00:05:36
    function. So it has to return an async
    00:05:39
    throwing stream. So we can use the for
    00:05:42
    tryw weight syntax here. So this is how
    00:05:45
    you create one of these and then you
    00:05:46
    have a continuation and you can yield
    00:05:49
    values into the continuation to have a
    00:05:51
    new entry in the stream. This is very
    00:05:54
    similar to how combine was used back in
    00:05:57
    the day. In there we create a task
    00:05:59
    because this is an async throwing
    00:06:01
    stream. We have some safety checks that
    00:06:04
    our speech recognizer is already created
    00:06:07
    and that it is available on the device.
    00:06:11
    And then this is basically the
    00:06:13
    interesting part. And as you can see,
    00:06:14
    it's actually not that much code.
    00:06:15
    There's just quite a bit of overhead and
    00:06:17
    some error handling in here. The
    00:06:20
    interesting thing is to do the speech
    00:06:22
    transcription, we need a speech engine
    00:06:25
    and a speech transcription request. And
    00:06:29
    we create these in the prepare engine
    00:06:31
    function just to keep this code a bit
    00:06:33
    more organized. So let's have a look at
    00:06:35
    the prepare engine function. And you can
    00:06:36
    already see this returns an AV audio
    00:06:39
    engine and then SF speech audio buffer
    00:06:42
    recognition request. So quite a
    00:06:44
    mouthful. And then this is basically
    00:06:46
    just a bunch of setup code for an audio
    00:06:48
    session. Making sure to set the category
    00:06:51
    to record with a measurement mode and
    00:06:54
    duck others options to make sure that
    00:06:56
    the audio is clean and usable for our
    00:06:59
    use case here. Then we create our audio
    00:07:01
    engine. We do a bit more setup here, but
    00:07:04
    that's not too interesting, I believe.
    00:07:06
    Second thing that we return aside from
    00:07:08
    our engine is our recognition request or
    00:07:11
    speech recognition request. So, we
    00:07:14
    create that over here.
    00:07:17
    It's important that we set should report
    00:07:18
    partial results to true because that's
    00:07:20
    what we want to do. We want to stream in
    00:07:23
    the um audio. We also want to add
    00:07:25
    punctuation because they the user might
    00:07:28
    say multiple sentences and use periods
    00:07:31
    or question marks for example. And we're
    00:07:33
    also going to tell the audio recognition
    00:07:36
    request that this is in fact a
    00:07:37
    dictation. This just helps it internally
    00:07:39
    to be a bit more accurate. If we want to
    00:07:41
    we can set requires ondevice recognition
    00:07:43
    to true to only use the ondevice models.
    00:07:46
    But in our use case it doesn't really
    00:07:48
    matter. All right. Then we do some more
    00:07:50
    setup. We prepare our engine. then we
    00:07:52
    start it and we return our engine and
    00:07:55
    our SF speech audio buffer recognition
    00:07:57
    request. So once we have called prepare
    00:07:59
    engine here, we are now locally uh
    00:08:02
    storing our audio engine and our request
    00:08:05
    and then we're creating our real
    00:08:07
    recognition task here. And once again
    00:08:10
    there's a lot of uh state checking and
    00:08:13
    error handling up here but actually it's
    00:08:16
    pretty simple.
    00:08:18
    So with our recognition task we get a
    00:08:20
    result and a an optional error here.
    00:08:25
    So if we have an error of course we uh
    00:08:28
    show some error state and we reset the
    00:08:30
    speech to text service. Not really
    00:08:32
    important right now. You can of course
    00:08:34
    change the implementation for your own
    00:08:36
    app. What's interesting is that we now
    00:08:39
    use the continuation that I mentioned
    00:08:41
    beforehand from our async throwing
    00:08:43
    stream. So we can iterate over these
    00:08:46
    results and we yield the best
    00:08:49
    transcription
    00:08:51
    as a formatted string. This works
    00:08:54
    because we set should report partial
    00:08:56
    results to true. And just as a reminder,
    00:08:58
    we get this transcription here within
    00:09:01
    the closure of our recognition task.
    00:09:03
    This result object is actually of type
    00:09:07
    SF speech recognition result and it
    00:09:09
    doesn't only have a best transcription.
    00:09:11
    It also has an is final property. So if
    00:09:15
    the transcription is final, we will
    00:09:17
    finish our continuation. So the for loop
    00:09:20
    will then stop and we will also reset
    00:09:23
    our speechtoext service to be clean
    00:09:26
    again for the next dictation. Then of
    00:09:28
    course we also have a stop transcribing
    00:09:30
    function and a reset function. You can
    00:09:32
    look into these yourself if you're
    00:09:34
    interested. Once again, this file is
    00:09:36
    linked as a GitHub gist in the video
    00:09:38
    description. And there we have it. We
    00:09:40
    have successfully added dictation to our
    00:09:42
    local AI chat app only using ondevice
    00:09:46
    firstparty Apple framework. So this is
    00:09:49
    actually super easy. SF speech analyzer
    00:09:52
    also works pretty accurately as far as
    00:09:55
    my tests are concerned and the community
    00:09:57
    feedback that I've heard. This is a big
    00:09:59
    improvement over the old APIs that we've
    00:10:02
    had previously. Of course, SF Speech
    00:10:04
    Analyzer is only available in iOS 26 and
    00:10:07
    up, but that's totally fine for our use
    00:10:09
    case as we're using Foundation models,
    00:10:11
    which is also only available in iOS 26
    00:10:13
    and

    Description:

    This lesson walks through an example of integrating SFSpeechRecognizer from Apple's Speech framework into an iOS 26 local AI chat app. Example Code: https://gist.github.com/chFlorian/0bc373278b11cff5ea547d112a5b5ac6 Join this channel to get access to perks: https://www.youtube.com/channel/UCYt_AtiKPyda44NYzwABvQQ/join 🚀 LaunchBuddy: https://apple.co/3iFcjjW 📚 Try CWC+: https://bit.ly/cwc_flo 🔭 Astro for ASO: https://flowritesco.de/astro ☕️ Buy me a coffee: https://ko-fi.com/flowritescode 👋 Links: https://flowritesco.de 🛠 Forge: https://apple.co/3riG8MQ Affiliate Links ❤ 📕 SwiftUI & Combine Books: https://www.bigmountainstudio.com/a/tpgmp 🔬 Get Reports about your apps: https://appfigures.com/r/5by3g 📊 Privacy focused analytics: https://dashboard.telemetrydeck.com/registration/organization?referralCode=27AOWO4R1TTEJBST 💻 The most powerful mac app for developers: https://devutils.app/?ref=flo ☕️ Support me: https://ko-fi.com/flowritescode If you have any video suggestions please feel free to let me know by a comment. Get in contact via Twitter: https://twitter.com/FloWritesCode

    Mediafile available in formats

    popular icon
    Popular
    hd icon
    HD video
    audio icon
    Only sound
    total icon
    All
    Show formatsreload icon
    * — If the video is playing in a new tab, go to it, then right-click on the video and select "Save video as..."
    ** — Link intended for online playback in specialized players

    Questions about downloading video

    question iconHow can I download "Transcribe Text via SFSpeechRecognizer (Lesson 05)" video?arrow icon

      http://univideos.ru/ website is the best way to download a video or a separate audio track if you want to do without installing programs and extensions.

      The UDL Helper extension is a convenient button that is seamlessly integrated into YouTube, Instagram and OK.ru sites for fast content download.

      UDL Client program (for Windows) is the most powerful solution that supports more than 900 websites, social networks and video hosting sites, as well as any video quality that is available in the source.

      UDL Lite is a really convenient way to access a website from your mobile device. With its help, you can easily download videos directly to your smartphone.

    question iconWhich format of "Transcribe Text via SFSpeechRecognizer (Lesson 05)" video should I choose?arrow icon

      The best quality formats are FullHD (1080p), 2K (1440p), 4K (2160p) and 8K (4320p). The higher the resolution of your screen, the higher the video quality should be. However, there are other factors to consider: download speed, amount of free space, and device performance during playback.

    question iconWhy does my computer freeze when loading a "Transcribe Text via SFSpeechRecognizer (Lesson 05)" video?arrow icon

      The browser/computer should not freeze completely! If this happens, please report it with a link to the video. Sometimes videos cannot be downloaded directly in a suitable format, so we have added the ability to convert the file to the desired format. In some cases, this process may actively use computer resources.

    question iconHow can I download "Transcribe Text via SFSpeechRecognizer (Lesson 05)" video to my phone?arrow icon

      You can download a video to your smartphone using the website or the PWA application UDL Lite. It is also possible to send a download link via QR code using the UDL Helper extension.

    question iconHow can I download an audio track (music) to MP3 "Transcribe Text via SFSpeechRecognizer (Lesson 05)"?arrow icon

      The most convenient way is to use the UDL Client program, which supports converting video to MP3 format. In some cases, MP3 can also be downloaded through the UDL Helper extension.

    question iconHow can I save a frame from a video "Transcribe Text via SFSpeechRecognizer (Lesson 05)"?arrow icon

      This feature is available in the UDL Helper extension. Make sure that "Show the video snapshot button" is checked in the settings. A camera icon should appear in the lower right corner of the player to the left of the "Settings" icon. When you click on it, the current frame from the video will be saved to your computer in JPEG format.

    question iconHow do I play and download streaming video?arrow icon

      For this purpose you need VLC-player, which can be downloaded for free from the official website https://www.videolan.org/vlc/.

      How to play streaming video through VLC player:

      • in video formats, hover your mouse over "Streaming Video**";
      • right-click on "Copy link";
      • open VLC-player;
      • select Media - Open Network Stream - Network in the menu;
      • paste the copied link into the input field;
      • click "Play".

      To download streaming video via VLC player, you need to convert it:

      • copy the video address (URL);
      • select "Open Network Stream" in the "Media" item of VLC player and paste the link to the video into the input field;
      • click on the arrow on the "Play" button and select "Convert" in the list;
      • select "Video - H.264 + MP3 (MP4)" in the "Profile" line;
      • click the "Browse" button to select a folder to save the converted video and click the "Start" button;
      • conversion speed depends on the resolution and duration of the video.

      Warning: this download method no longer works with most YouTube videos.

    question iconWhat's the price of all this stuff?arrow icon

      It costs nothing. Our services are absolutely free for all users. There are no PRO subscriptions, no restrictions on the number or maximum length of downloaded videos.