How to detect long utterances use Yamnet VAD

[developerzjy]

hi, I want to detect long utterances use Yamnet VAD, but I can't find the api configuration. My purpose is to detect if someone speaks in a period of time, such as within 10 seconds someone said a word. I don't care about the result of each frame, I just want a final result. Does Yamnet support this case？what should I do？

[gkonovalov]

Hi @developerzjy

Yes, Yamnet VAD support continuous speech recognition, please see code below:

val vad = Vad.builder()
        .setContext(applicationContext)
        .setSampleRate(SampleRate.SAMPLE_RATE_16K)
        .setFrameSize(FrameSize.FRAME_SIZE_243)
        .setMode(Mode.NORMAL)
        .setSilenceDurationMs(30)
        .setSpeechDurationMs(30)
        .build()

    vad.setContinuousClassifierListener("Speech", audioData, object : VadListener {
        override fun onResult(event: SoundCategory) {
            when (event.label) {
                "Speech" -> "Speech Detected!" + event.score
                else -> "Noise Detected!" + event.score
            }
        }
    })

    vad.close()

Full code example: https://github.com/gkonovalov/android-vad/blob/main/example/src/main/java/com/konovalov/vad/example/VadYamnetFragment.kt

[developerzjy]

I saw this code and run it, I need additional code to convert the result of each frame into a final result.
thank you