Streaming TTS - RunAnywhere Documentation

synthesizeStream emits AudioChunk values as they are produced, so playback starts on the first chunk instead of the last.

RunAnywhere.tts.synthesizeStream(longText).collect { chunk ->
    player.enqueue(chunk)
}

Signature

fun synthesizeStream(text: String, options: TtsOptions? = null): Flow<AudioChunk>

When it is worth it

For a sentence, speak is simpler and the latency difference is invisible. For a paragraph, or for a model reply being streamed as it generates, streaming synthesis is what keeps the gap between “the model finished” and “the user hears something” short.

Pairing it with a streaming completion

Speak each sentence as the model finishes it:

val sentence = StringBuilder()

RunAnywhere.llm.generateStream(prompt).collect { event ->
    if (event !is GenerationEvent.Token) return@collect
    sentence.append(event.text)

    if (event.text.any { it in ".!?" }) {
        RunAnywhere.tts.speak(sentence.toString())
        sentence.clear()
    }
}

if (sentence.isNotEmpty()) RunAnywhere.tts.speak(sentence.toString())

For a full conversation rather than one reply, use the voice session, which handles turn-taking and interruption for you.

Feeding AudioTrack

val track = AudioTrack.Builder()
    .setAudioFormat(
        AudioFormat.Builder()
            .setEncoding(AudioFormat.ENCODING_PCM_FLOAT)
            .setSampleRate(22050)
            .setChannelMask(AudioFormat.CHANNEL_OUT_MONO)
            .build()
    )
    .setTransferMode(AudioTrack.MODE_STREAM)
    .build()

track.play()

RunAnywhere.tts.synthesizeStream(text).collect { chunk ->
    track.write(chunk.samples, 0, chunk.samples.size, AudioTrack.WRITE_BLOCKING)
}

track.stop()
track.release()

Release the track when you are done. It holds an audio output.