Speaker

Overview

Maverick AI glasses have a built-in speaker. The SDK handles it via the M2AudioService (accessed from Evs.audioService).

There are two ways to play audio on the glasses:

Way	Use it for	API
Stream	Live audio: AI / text-to-speech replies, media, internet radio	`openStream` → `write` → `endStream` / `closeStream`
Sound resource	Short sounds you play again and again: chimes, alerts, UI feedback	`M2AudioResource` + `playSound`

The SDK converts your audio to the glasses' own format (LC3) and sends it over BLE at real-time pace.

Streaming Audio

Open a stream with the format of the bytes you will pass to write:

Input	Description
`M2AudioStreamInput.Pcm16(sampleRate, channels)`	Raw signed 16-bit little-endian PCM, mono or stereo, any sample rate
`M2AudioStreamInput.Encoded(format)`	Encoded audio - MP3, AAC, M4A or WAV. The SDK decodes it.
`M2AudioStreamInput.Lc3Frames`	Audio already in the glasses' LC3 format, forwarded as is
`M2AudioStreamInput.Url(url)`	An internet radio stream (ICY / HTTP MP3 or AAC). The SDK reads it itself - do not call `write`.

Only one stream can be open at a time.

KotlinSwift

// Text-to-speech output: 24 kHz mono PCM
Evs.audioService.openStream(M2AudioStreamInput.Pcm16(sampleRate = 24000))

Evs.audioService.write(pcmChunk)      // call as audio arrives
// ...
Evs.audioService.endStream()          // after the last chunk: let the tail finish playing

// Text-to-speech output: 24 kHz mono PCM
let audio = Evs.shared.audioService
audio.openStream(input: M2AudioStreamInput.Pcm16(sampleRate: 24000, channels: .mono))

audio.write(data: pcmChunk, offset: 0, size: pcmChunk.size)   // call as audio arrives
// ...
audio.endStream()                                             // let the tail finish playing

Ending a Stream

Call	What it does
`endStream()`	Graceful end: plays the audio still buffered, then closes. Use after the last `write`.
`closeStream()`	Stops now and drops anything still buffered. Use to cancel.
`flush()`	Drops the buffered audio but keeps the stream open - for when the user interrupts the AI mid-sentence and a new reply starts right away.

For Encoded streams you do not need endStream(): the SDK reports onAudioStreamFinished by itself when the decoder reaches the end of the input and the speaker has played it.

Stream Events

Kotlin

Evs.audioService.registerStreamListener(object : M2AudioService.IOnAudioStreamEvents {
    override fun onAudioStreamChanged(input: M2AudioStreamInput?) { /* opened / closed */ }
    override fun onAudioStreamStarted(input: M2AudioStreamInput) { /* first audio reached the glasses */ }
    override fun onAudioStreamFinished(input: M2AudioStreamInput) { /* Encoded stream fully played */ }
    override fun onAudioStreamError(error: M2AudioStreamError) { /* error.code, error.message */ }
    override fun onStreamMetadata(title: String) { /* internet radio: now playing */ }
})

write returns false when the SDK rejects the input straight away (no stream open, wrong data for the format). getLastStreamError() returns the reason.

Sound Resources

A sound is uploaded to the glasses once and played from there, so replaying it costs no BLE traffic. Supported source formats: WAV, MP3, AAC, M4A, and LC3.

KotlinSwift

// From a file bundled with the app - the format comes from the extension
val chime = M2AudioResource("sounds/chime.wav")

Evs.audioService.playSound(chime)                 // uploads on first use, then plays
Evs.audioService.playSound(chime, repeat = true)  // loop
Evs.audioService.stopSound(chime)

let chime = M2AudioResource(nameWithExtension: "sounds/chime.wav", persistent: false, tag: nil)

Evs.shared.audioService.playSound(audio: chime, repeat: false)
Evs.shared.audioService.stopSound(audio: chime)

In-memory audio works too: M2AudioResource(bytes, M2AudioFormat.Mp3).

Call	What it does
`playSound(audio, repeat)`	Plays the sound, uploading it first if needed
`playSoundAdvanced(audio, repeat, offsetMillis, durationMillis)`	Plays part of the sound
`stopSound(audio)`	Stops it
`releaseSound(audio)`	Frees its slot on the glasses. A `persistent` sound stays cached for next time.
`deleteSound(audio)`	Removes it from the glasses, cache included

Pass persistent = true when creating the resource to keep the sound in the glasses' cache across sessions.

Volume

Call	Description
`setVolume(volume)` / `getVolume()`	Speaker volume, 0-100. Default 80.
`setDigitalGain(gain)`	Extra software boost above the hardware maximum. `1.0` = off; up to about `4.0` (+12 dB). Costs CPU - use only as much as needed.

Notes

Close the stream when you are done - an open stream keeps the speaker on.
With the Glasses Simulator, the audio plays on the PC speaker.

See Also

Microphone - Capture audio from the glasses microphones

Resources - Upload and cache resources on the glasses