Skip to content

Missing word confidence + unclear text confidence value #27

@Megamannen

Description

@Megamannen

When using setMaxAlternatives the confidence level from words are gone and the confidence level on the whole text is very strange number.

I was expecting a value between 0.0 and 1.0, but instead I get numbers in the hundreds without any clue on what they represent

  "alternatives" : [{
      "confidence" : 297.045959, // of what ? _ ?
      "result" : [{
          "end" : 2.130000,
          "start" : 1.740000,
          "word" : "i'm",
          // "conf": "IS GONE"
        }, {
          "end" : 2.160000,
          "start" : 2.130000,
          "word" : "a"
          // "conf": "IS GONE"
        }, {
          "end" : 2.640000,
          "start" : 2.160000,
          "word" : "guy"
          // "conf": "IS GONE"
        }],
      "text" : "i'm a guy"
    }]

Can be reproduced with modified example

<!DOCTYPE html>
<html>
  <head>
    <script src="https://cdn.jsdelivr.net/gh/msqr1/[email protected]/Examples/Vosklet.js" async defer></script>
    <script>
      async function start() {
        // All data is collected and transfered to the main thread so the AudioContext won't output anything. Set sinkId type to none to save power
        let ctx = new AudioContext({sinkId: {type: "none"}});

        // Setup microphone   
        let micNode = ctx.createMediaStreamSource(await navigator.mediaDevices.getUserMedia({
          video: false,
          audio: {
            echoCancellation: true,
            noiseSuppression: true,
            channelCount: 1
          },
        }));
        
        let module = await loadVosklet();

        let model = await module.createModel("https://ccoreilly.github.io/vosk-browser/models/vosk-model-small-en-us-0.15.tar.gz","English","vosk-model-small-en-us-0.15");        
        let recognizer = await module.createRecognizer(model, ctx.sampleRate);
        recognizer.setWords(true);
        recognizer.setMaxAlternatives(2);

        // Listen for result and partial result
        recognizer.addEventListener("result", ev => console.log("Result: ", ev.detail));
        recognizer.addEventListener("partialResult", ev => console.log("Partial result: ", ev.detail));

        // Create a transferer node to get audio data on the main thread
        let transferer = await module.createTransferer(ctx, 128 * 150);

        // Recognize data on arrival
        transferer.port.onmessage = ev => {
          recognizer.acceptWaveform(ev.data)
        };

        // Connect transferer to microphone
        micNode.connect(transferer);
      }
    </script>
    <!-- Start and create audio context only as a result of user's action -->
    <button onclick="start()">Start</button>
  </head>
</html>

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions