Add live captions to a video call app with daily-js

This page summarizes the projects mentioned and recommended in the original post on

Our great sponsors
  • Appwrite - The Open Source Firebase alternative introduces iOS support
  • Scout APM - Less time debugging, more time building
  • SonarQube - Static code analysis for 29 languages.
  • prebuilt-transcription

    If you are like spoilers and want to see what we’re building today, you can jump straight into the prebuilt-transcription code and also try a live demo

  • react-window

    React components for efficiently rendering large lists and tabular data

    Without this styling for the and its container, Daily Prebuilt defaults to taking up only a small amount of space on the page. You can change the styling however you like and Daily Prebuilt will fit within those constraints.

    Note: We are using wrap() to add Daily Prebuilt to an existing , but you could also use the createFrame method to make a new , style that frame, and add it to the page.

    Video call UI when a token or call owner is not present and transcription can't be used

    Transcription component

    Now that we have Daily Prebuilt loaded on the page, let’s start implementing transcription by adding a component to store our buttons and transcript.

    From our [room].tsx, we reference the Transcription component and pass some managed state to the component:

    • callFrame: passes the Daily call frame object, which allows the component to start and stop transcription
    • newMsg: sends each new transcripted message to the component for showing the text in the transcript window
    • owner: this boolean tells the component whether the current user is or isn’t a room owner
    • isTranscribing: this boolean tells the component that Daily is or isn’t currently transcribing.
    Enter fullscreen mode Exit fullscreen mode

    In our Transcription component (defined in components/Transcription.tsx), we have a button that toggles the option to start or stop transcription based on whether transcription is currently active according to Daily (we’ll come back to that in a second):

     onClick={() => {
       isTranscribing ? stopTranscription() : startTranscription();
     {isTranscribing ? "Stop transcribing" : "Start transcribing"}
    Enter fullscreen mode Exit fullscreen mode

    If the meeting participant is not an owner, this button will be disabled along with a message explaining that only meeting room owners can start transcription.

    This button utilizes these two simple functions:

     async function startTranscription() {
     async function stopTranscription() {
    Enter fullscreen mode Exit fullscreen mode

    How do these functions know if transcription is happening or not? For that, we jump back to [room].tsx. Earlier in the post, we looked at the basic structure of the startCall function. In our demo, this function also has a few lines dedicated to Daily event listeners. We are listening to a few Daily-emitted events that help us shape the video call experience. Two of these events are transcription-started and transcription-stopped events.

    When those events are emitted, we know to update the React state to set isTranscribing to its correct boolean value.

       newCallFrame.on("transcription-started", () => {
       newCallFrame.on("transcription-stopped", () => {
    Enter fullscreen mode Exit fullscreen mode

    Note: You can also use our new Daily React Hooks library to more quickly connect your React-based app with Daily’s JavaScript API!

    Adding transcription

    Now that we are able to start and stop transcription, we need to add the transcripts to the page. Our transcripts come in from Daily via an ”app-message” event. For that, we need another event listener within our startCall function. This checks whether each ”app-message” came from the ID of “transcription” and whether it is a full sentence (that’s what data.is_final is doing below). When we have a message, we save the message as an object with the author’s username, the text transcription, and a timestamp.

         (msg: DailyEventObjectAppMessage | undefined) => {
           const data = msg?.data;
           if (msg?.fromId === "transcription" && data?.is_final) {
             const local = newCallFrame.participants().local;
             const name: string =
               local.session_id === data.session_id
                 ? local.user_name
                 : newCallFrame.participants()[data.session_id].user_name;
             const text: string = data.text;
             const timestamp: string = data.timestamp;
             if (name.length && text.length && timestamp.length) {
               setNewMsg({ name, text, timestamp });
    Enter fullscreen mode Exit fullscreen mode

    We need some React state to hold messages, so we set up a const where we instantiate this state as an empty array to hold incoming message objects.

    const [messages, setMessage] = useState<Array<transcriptMsg>>([]);
    Enter fullscreen mode Exit fullscreen mode

    This is essentially all that needs to be done to get transcription on the page. You can loop through this array of messages and add them to the screen, or you can add each new message to the screen as it arrives. However, there’s one extra step worth taking to optimize your app for all of these messages, and we’ll see how that works in the next section.

    Note: Transcript messages are ephemeral. They are only available for the message the user has received while they are in the room. If you refresh your page, you’ll lose the transcripts. Similarly, new users will only see a transcript for conversations that have taken place since they’ve joined and not a history.

    Transcribed text scrolling up and down

    Optimize your window

    Seeing transcripts appear on the screen is super fun, but it can quickly slow down browser windows with the addition of so many DOM elements getting added to the screen. Below, we’ll cover not just how we add transcript messages to our page, but also how to do it in a way that is efficient and not overwhelming to anyone’s browser.

    To help with this, we need to add two dependencies to our app: react-window and react-virtualized-auto-sizer. These libraries help us by loading only the most recent messages. Instead of loading the entire array of message objects as HTML, the DOM only loads the small part of the data set visible in the window. This virtualization technique prevents poor performance caused by an overloaded browser tab holding too much data in memory. Users can still scroll up and see previous messages which are loaded as needed when requested.

    We have established consts for the transcript list and rows that instantiate as empty objects.

     const listRef = useRef<any>({});
     const rowRef = useRef<any>({});
    Enter fullscreen mode Exit fullscreen mode

    We add new messages received from the parent [room] page to an array. We also have a small function that keeps the array of messages moving to the bottom (most recent) element every time a message is received.

     useEffect(() => {
       setMessage((messages: Array<transcriptMsg>) => {
         return [...messages, newMsg];
     }, [newMsg]);
     useEffect(() => {
       if (messages && messages.length > 0) {
         return () => {
     }, [messages]);
    Enter fullscreen mode Exit fullscreen mode

    For each row, we call a formatting function. It structures the transcript in the style of “Message Author: Message Text” on the left and Timestamp trimmed to a local time only on the right (styled with CSS, the handy-dandy float:right).

     function Row({ index, style }: rowProps) {
       const rowRef = useRef<any>({});
       useEffect(() => {
         if (rowRef.current) {
           setRowHeight(index, rowRef.current.clientHeight);
       }, [rowRef]);
       return (
         <div style={style}>
           {messages[index].name && (
             <div ref={rowRef}>
               {messages[index].name}: {messages[index].text}
               <span className={styles.timestamp}>
                 {new Date(messages[index].timestamp).toLocaleTimeString()}
    Enter fullscreen mode Exit fullscreen mode

    Our rendered transcript block then looks like this, with each loaded row wrapped in the react-window List and react-virtualized-auto-sizer AutoSizer elements.

     {({ height, width }) => (
    Enter fullscreen mode Exit fullscreen mode


    The transcripts collected in this app are not available after the call concludes, so downloading them is helpful if you want to use them later.

    To do that, we need to prepare a chat file with all of the text, not just the text currently virtualized on the screen.

    We have already seen that we are using React state to collect and set messages. For preparing a plain text file with the transcript inside, we will add a transcriptFile state that instantiates as an empty string.

    const [transcriptFile, setTranscriptFile] = useState<string>("");
    Enter fullscreen mode Exit fullscreen mode

    Next, let’s set up a useEffect to style the transcript in a way that works best for reviewing later. Unlike the live transcript where we have the timestamp on the right and set to local time only, this includes the full timestamp and date for every message.

       Allow user to download most recent full transcript text
     useEffect(() => {
           new Blob(
                 ? `${msg.timestamp} ${}: ${msg.text}\n`
                 : `Transcript\n`
             { type: "octet/stream" }
     }, [messages]);
    Enter fullscreen mode Exit fullscreen mode

    This link will get the most recent full transcript and by default save it as a file called transcript.txt, although this can be changed later by the user.

    <a href={transcriptFile} download="transcript.txt">
      Download Transcript
    Enter fullscreen mode Exit fullscreen mode


    And there you have it! Using Daily Prebuilt and our new Transcription API with Deepgram, it’s not too much work to add a live transcript to your meetings. From what we’ve shown in this demo, you can easily add different styles (including to the Daily Prebuilt window itself by customizing with your own color themes)

    We would love to see what you’ve built using Daily. Reach out to us anytime at [email protected]!

  • Appwrite

    Appwrite - The Open Source Firebase alternative introduces iOS support . Appwrite is an open source backend server that helps you build native iOS applications much faster with realtime APIs for authentication, databases, files storage, cloud functions and much more!

  • Next.js

    The React Framework

    This demo is based on the Next.js React framework, starting with the create-next-app template builder. This tutorial also uses TypeScript. If you are new to TypeScript, no worries! Because it’s built on top of JavaScript, it looks very similar with a few additional features and syntax.

NOTE: The number of mentions on this list indicates mentions on common posts plus user suggested alternatives. Hence, a higher number means a more popular project.

Suggest a related project

Related posts