Voice Agent

val session = RunAnywhere.voice.createSession(
    stt = ModelRef("whisper-base"),
    llm = ModelRef("qwen3-0.6b"),
    tts = ModelRef("piper-en-amy")
)

session.start()

session.events.collect { event ->
    when (event) {
        is VoiceEvent.UserTranscript -> transcript += "user: ${event.text}"
        is VoiceEvent.AssistantText -> transcript += "assistant: ${event.text}"
        else -> {}
    }
}

suspend fun createSession(
    stt: ModelRef,
    llm: ModelRef,
    tts: ModelRef,
    vad: VadOptions? = null,
    turnHandling: TurnHandlingOptions? = null,
    generation: LlmOptions? = null,
    downloadIfNeeded: Boolean = true
): VoiceSession

val events: Flow<VoiceEvent>
fun start()
fun say(text: String): SpeechHandle
suspend fun interrupt()
fun close()

val session = RunAnywhere.voice.createSession(
    stt = sttModel,
    llm = llmModel,
    tts = ttsModel,
    turnHandling = TurnHandlingOptions(
        endpointing = Endpointing(minDelayMs = 500, maxDelayMs = 3000),
        interruption = Interruption(enabled = true, minDurationMs = 500)
    )
)

generation = LlmOptions(
    systemPrompt = "You are a voice assistant. Answer in one or two sentences.",
    maxOutputTokens = 100
)

class VoiceAgentViewModel(app: Application) : AndroidViewModel(app) {
    private val _transcript = MutableStateFlow<List<Pair<String, String>>>(emptyList())
    val transcript = _transcript.asStateFlow()

    private var session: VoiceSession? = null

    fun start() {
        viewModelScope.launch {
            val s = RunAnywhere.voice.createSession(
                stt = ModelRef("whisper-base"),
                llm = ModelRef("qwen3-0.6b"),
                tts = ModelRef("piper-en-amy"),
                generation = LlmOptions(
                    systemPrompt = "Answer in one or two sentences.",
                    maxOutputTokens = 100
                )
            )
            session = s

            s.start()
            s.say("Hello. What can I help with?")

            s.events.collect { event ->
                when (event) {
                    is VoiceEvent.UserTranscript ->
                        _transcript.value += "user" to event.text
                    is VoiceEvent.AssistantText ->
                        _transcript.value += "assistant" to event.text
                    else -> {}
                }
            }
        }
    }

    override fun onCleared() {
        session?.close()
        session = null
        super.onCleared()
    }
}

if (ContextCompat.checkSelfPermission(context, Manifest.permission.RECORD_AUDIO)
    != PackageManager.PERMISSION_GRANTED) {
    launcher.launch(Manifest.permission.RECORD_AUDIO)
}

val result = RunAnywhere.diarization.diarize(AudioInput.file(path))