On-device Qwen3-TTS (12 Hz talker + MTP + codec) packaged by TheStage AI for Apple Silicon. Batch and push-based streaming via the TheStage Apple SDK; 24 kHz mono PCM out.


Runtime	TheStage Apple SDK (`Qwen3TTSPipeline`)
Sample rate	24 000 Hz mono
Default voice	`b_ref`
HF engines	`TheStageAI/Qwen3-TTS-12Hz-0.6B-Base`

Overview

This repo hosts CoreML engine bundles for Qwen3-TTS on the TheStage Apple SDK. start_model detects the MTP / talker layout and routes to Qwen3TTSPipeline automatically (vs NeuTTS).

System Requirements

Property	Value
Hardware	Apple Silicon Mac, or physical iPhone / iPad
macOS	15.0+
iOS	18.0+
Xcode	16.0+
Swift	6.0+
Flutter (optional)	3.24+

Simulator is not supported — run on real Apple Silicon hardware.

TheStage AI Access Token

Create a token at app.thestage.ai and pass it to the SDK:

try await TheStageAI.shared.initialize(apiToken: "th_…")

await TheStageFlutterSDK.initialize(api_token: 'th_…');

Token is checked online in initialize (once per app process when reachable). Inference runs fully on-device. Offline initialize fails — reconnect and call initialize again.

TheStage Apple SDK

Docs: TheStage Apple SDK · TTS · Repo: TheStage Apple SDK

Installation (SwiftPM)

In Xcode: File → Add Package Dependencies…, paste https://github.com/TheStageAI/AppleSDK.git, and add the TheStageSDK product. Or in Package.swift:

swift

.package(
    url: "https://github.com/TheStageAI/AppleSDK.git",
    exact: Version(1, 1, 0)
)

Installation (Flutter / iOS)

yaml

# pubspec.yaml
dependencies:
  thestage_apple_sdk:
    git:
      url: https://github.com/TheStageAI/AppleSDK.git
      path: plugin/thestage_apple_sdk
      ref: 1.1.0

Swift — batch

swift

import TheStageSDK

try await TheStageAI.shared.initialize(apiToken: "th_…")

let tts = try await Qwen3TTSPipeline(
    engines_path: "TheStageAI/Qwen3-TTS-12Hz-0.6B-Base",
    voice_id: "b_ref"
)

let result = tts.infer(text: "Hello, world!")
// result.samples: [Float] @ 24 kHz mono

Swift — streaming

Same push pattern as NeuTTS: open_streamer() → drain output concurrently → send(...) → stop_stream().

Flutter (iOS)

dart

await TheStageFlutterSDK.start_model(
  model_name: 'tts',
  engines_path: 'TheStageAI/Qwen3-TTS-12Hz-0.6B-Base',
  config: {'voice_id': 'b_ref'},
);

final result = await TheStageFlutterSDK.infer(
  model_name: 'tts',
  input_json: {'text': 'Hello, world!'},
);

Sampling defaults come from the bundle (talker typically ~temperature=0.9, top_k=50); override via TTSGenerationConfig / JSON config.

On-device latency

Release build on Apple M2 Max (NPU / ANE), macOS 26.2 (guidance, not an SLA):

Metric	Value
Batch rtf	1.76
Stream rtfx	2.18
TTFA (s)	0.30
Process mem (MB)	~423

Acknowledgments

This work builds on Qwen3-TTS-12Hz-0.6B-Base by Qwen / Alibaba: Qwen/Qwen3-TTS-12Hz-0.6B-Base.

This Apple Silicon package is produced by TheStage AI; it is not an official Qwen release.

TheStageAI/Qwen3-TTS-12Hz-0.6B-Base

Qwen3-TTS 12Hz 0.6B Base (Apple Silicon)