macOS Expression 2

Play with sound (8 seconds, captions)

The shortest native Apple path: a command-line tool that opens an avatar, feeds it a 16 kHz WAV and pulls 416×720 frames, rendered on this Mac; the engine contacts bitHuman only to check your API secret and report session time. It writes the first frame to out/first-frame.png; the clip above muxes every frame it pulled with the audio.

Requirements

Get the code

git clone https://github.com/bithuman-product/bithuman-examples.git
cd bithuman-examples/swift/macos-expression2
./setup.sh

setup.sh downloads the wise-pup avatar, the shared engine files and a speech clip into Model/. For your own avatar run BITHUMAN_API_SECRET=… ./setup.sh <AGENT_CODE>.

Set your API secret

export BITHUMAN_API_SECRET="<your API secret>"

Run it

swift run -c release MacOSExpression2

Expected output

engine ready: 416x720, isReady=true
audio: 325451 samples, 20.34 s
generated 407 frames in … s (… FPS, …x real time) -> out/first-frame.png

407 frames for 20.34 seconds of audio is 20 fps. The first run prepares the engine for your Mac; keep Model/staged/ and later runs start faster. The run prints per-chunk log lines between the three lines above; they are harmless.

How it works

Sources/main.swift is about 60 lines around three calls from the Swift package’s Expression2 product:

let engine = try Expression2Engine.create(
    avatarContainer: model.appendingPathComponent("agent.imx"),
    sharedEngineContainer: model.appendingPathComponent("shared-engine.imx"),
    stagingDir: model.appendingPathComponent("staged"))
engine.feed(samples)      // 16 kHz mono float, any length
engine.flushTail()        // end of the utterance
while let (frame, _) = engine.pull() { /* 416×720 BGR, 3 bytes per pixel */ }

pull() returns nil until a chunk of frames is ready, so the example polls until it has drained them all. The API is on Apple.

The code that matters

The whole render path of Sources/main.swift: open the avatar, feed the utterance, drain the frames.

// excerpt: swift/macos-expression2/Sources/main.swift
// 1. Open the identity. Both downloads are containers, opened as they are;
//    `stagingDir` is a writable directory the engine unpacks them into once.
//    Keep it between runs and the next start is much faster.
let engine = try Expression2Engine.create(
    avatarContainer: model.appendingPathComponent("agent.imx"),
    sharedEngineContainer: model.appendingPathComponent("shared-engine.imx"),
    stagingDir: model.appendingPathComponent("staged"))
print("engine ready: \(engine.width)x\(engine.height), isReady=\(engine.isReady)")

let samples = try readPCM(model.appendingPathComponent("speech16k.wav"))
print("audio: \(samples.count) samples, "
    + String(format: "%.2f s", Double(samples.count) / 16_000))

// 2. Feed the whole utterance, then drain. Generation is asynchronous:
//    `pull()` returns nil until a chunk of frames lands, so a drain on the
//    line after `feed()` gets nothing at all. Poll — 100 idle ticks is done.
let started = Date()
engine.feed(samples)
engine.flushTail()

var frames = 0, idleTicks = 0
while idleTicks < 100 {
    var got = false
    while let (frame, _) = engine.pull() {
        if frames == 0 {
            writePNG(frame, width: engine.width, height: engine.height,
                     to: out.appendingPathComponent("first-frame.png"))
        }
        frames += 1
        got = true
    }
    if got { idleTicks = 0 } else { idleTicks += 1; usleep(50_000) }
}

The complete file is on GitHub.

Make it your own

• Your own avatar: create one with the Agents API ("model": "expression-2"), then BITHUMAN_API_SECRET=… ./setup.sh <AGENT_CODE>.
• Your own audio: any 16 kHz mono 16-bit WAV works as Model/speech16k.wav; afconvert -f WAVE -d LEI16@16000 -c 1 in.m4a Model/speech16k.wav converts one.
• A window instead of a file: the iOS example is the same engine in a SwiftUI app with a microphone button, and it also builds for iPad.

Troubleshooting

Next

• iOS example · iOS & iPadOS · CLI example · source on GitHub