speech-api/SpeechRecognition-phrases-manual.https.html

<!DOCTYPE html>
<html lang="en">
<meta name="timeout" content="long">
<title>SpeechRecognition Phrases</title>

<script src="/resources/testdriver.js"></script>
<script src="/resources/testdriver-vendor.js"></script>
<script src="/resources/testharness.js"></script>
<script src="/resources/testharnessreport.js"></script>

<script>
async function getAudioTrackFromFile(filePath) {
    const audioContext = new AudioContext();
    const response = await fetch(filePath);
    const arrayBuffer = await response.arrayBuffer();
    const audioBuffer = await audioContext.decodeAudioData(arrayBuffer);
    const source = audioContext.createBufferSource();
    source.buffer = audioBuffer;

    const destination = audioContext.createMediaStreamDestination();
    source.connect(destination);
    source.start();

    return destination.stream.getAudioTracks()[0];
}

promise_test(async (t) => {
    window.SpeechRecognition = window.SpeechRecognition || window.webkitSpeechRecognition;

    // Install en-US for on-device speech recognition.
    const installOptions = { langs: ["en-US"], processLocally: true };
    const installPromise = test_driver.bless(
        "Install on-device en-US speech recognition",
        () => SpeechRecognition.install(installOptions)
    );
    assert_true(
        installPromise instanceof Promise,
        "SpeechRecognition.install() should return a Promise."
    );
    const installResult = await installPromise;
    assert_true(
        installResult,
        "SpeechRecognition.install() for en-US should resolve with true."
    );

    // Verify the audio track for recognition context exists.
    const audioTrack = await getAudioTrackFromFile("/media/recognition_context.mp3");
    assert_true(
        audioTrack instanceof MediaStreamTrack,
        "Audio track should be a valid MediaStreamTrack"
    );

    // Create the first speech recognition with a mode that does not support contextual biasing.
    // Note that this may vary between browsers in the future.
    const recognition1 = new SpeechRecognition();
    recognition1.processLocally = false;
    recognition1.lang = "en-US";

    recognition1.onerror = function(event) {
        assert_equals(
            event.error,
            "phrases-not-supported",
            "First speech recognition should throw a phrases-not-supported error"
        );
    };

    recognition1.phrases = new SpeechRecognitionPhraseList([
        new SpeechRecognitionPhrase("test", 1.0)
    ]);

    // Create the second speech recognition with a mode that supports contextual biasing.
    const recognition2 = new SpeechRecognition();
    recognition2.processLocally = true;
    recognition2.lang = "en-US";

    recognition2.onerror = function(event) {
        // Currently WPT may not be able to detect that SODA is available and
        // will throw a "language-not-supported" error here.
        assert_unreached("Caught an error: " + event.error);
    };

    recognition2.phrases = new SpeechRecognitionPhraseList([
        new SpeechRecognitionPhrase("ASIC", 1.0),
        new SpeechRecognitionPhrase("FPGA", 1.0)
    ]);

    const recognitionPromise = new Promise((resolve) => {
        recognition2.onresult = (event) => {
            const transcript = event.results[0][0].transcript;
            const words = transcript.toLowerCase().split(' ');
            // Resolve when the last word is "expectations".
            if (words.length > 0 && words[words.length - 1] === "expectations") {
                resolve(transcript);
            }
        };
    });
    recognition2.start(audioTrack);

    const transcript = await recognitionPromise;
    assert_equals(
        transcript.toLowerCase(),
        "the report confirmed that the asic's throughput and " +
        "the fpga's latency were both below expectations",
        "Second speech recognition should correctly recognize the phrases"
    );
}, "SpeechRecognition should recognize speech with the given contextual information.");
</script>
</html>