Sending audio data

To recognize PCM audio supplied by the app instead of using the vehicle microphone, pass a ByteArray to sendAudio(). This feature is available only in ON_DEVICE mode. You can send the entire buffer at once or stream it in multiple chunks.

info

The 3,200-byte chunk and 100 ms interval in this document are an example of streaming 16 kHz, 16-bit, mono PCM at real-time speed. For another audio format, calculate the chunk size that corresponds to 100 ms of that format.

Send a complete buffer

Kotlin

val pcmData = applicationContext.assets
    .open("test_audio.pcm")
    .use { it.readBytes() }

speechToText.sendAudio(pcmData)
speechToText.completeAudioSend()

Java

byte[] pcmData;
try (InputStream input = getAssets().open("test_audio.pcm");
     ByteArrayOutputStream output = new ByteArrayOutputStream()) {
    byte[] buffer = new byte[8_192];
    int read;
    while ((read = input.read(buffer)) != -1) {
        output.write(buffer, 0, read);
    }
    pcmData = output.toByteArray();
}

speechToText.sendAudio(pcmData);
speechToText.completeAudioSend();

Stream chunks

Kotlin

private const val BYTES_PER_100_MS = 3_200

val pcmData = applicationContext.assets
    .open("test_audio.pcm")
    .use { it.readBytes() }

pcmData.asList()
    .chunked(BYTES_PER_100_MS)
    .forEach { chunk ->
        speechToText.sendAudio(chunk.toByteArray())
        Thread.sleep(100L)
    }

speechToText.completeAudioSend()

Java

private static final int BYTES_PER_100_MS = 3_200;

byte[] buffer = new byte[BYTES_PER_100_MS];
try (InputStream input = getAssets().open("test_audio.pcm")) {
    int read;
    while ((read = input.read(buffer)) != -1) {
        speechToText.sendAudio(Arrays.copyOf(buffer, read));
        Thread.sleep(100L);
    }
}

speechToText.completeAudioSend();

After sending the final audio data, always call completeAudioSend() to signal that the input is complete. The previously registered ResultListener receives the recognition results.