Argmax Open-Source SDK Swift

The Argmax Open-Source SDK Swift is a collection of turn-key on-device inference frameworks for Apple platforms. It provides three primary kits:

• WhisperKit: Speech-to-text using OpenAI Whisper.
• SpeakerKit: Speaker diarization using Pyannote.
• TTSKit: Text-to-speech using Qwen-TTS.

Note: For advanced features like real-time transcription with speakers, custom vocabulary, or Android support, use the Argmax Pro SDK.

To use the SDK in your Swift package, add the dependency to your Package.swift and then specify the desired product in your target dependencies. You can use the ArgmaxOSS umbrella product or individual kits.

.target(
    name: "YourApp",
    dependencies: [
        // Import everything at once:
        .product(name: "ArgmaxOSS", package: "argmax-oss-swift"),

        // Or pick individual kits:
        // .product(name: "WhisperKit", package: "argmax-oss-swift"),   // speech-to-text
        // .product(name: "TTSKit", package: "argmax-oss-swift"),       // text-to-speech
        // .product(name: "SpeakerKit", package: "argmax-oss-swift"), // speaker diarization
    ]
),

The local server's OpenAPI specification can be used to generate clients for various languages.

Generate Python client:

swift run swift-openapi-generator generate scripts/specs/localserver_openapi.yaml \
  --output-directory python-client \
  --mode client \
  --mode types

Generate TypeScript client:

npx @openapitools/openapi-generator-cli generate \
  -i scripts/specs/localserver_openapi.yaml \
  -g typescript-fetch \
  -o typescript-client

The benchmark process uses Fastlane to pass the model name to Xcode via an environment variable. To ensure the environment is set up correctly:

1. Open the example project: xed Examples/WhisperAX.
2. Click on the WhisperAX scheme at the top of the Xcode window and select Edit Scheme.
3. Under Environment Variables, verify there is an entry with MODEL_NAME as the name and $(MODEL_NAME) as the value.

The Swift CLI is useful for testing transcription outside of an Xcode project.

Setup:

git clone https://github.com/argmaxinc/argmax-oss-swift.git
cd argmax-oss-swift
make setup
# Download a specific model (requires git-lfs)
make download-model MODEL=large-v3-v20240930_626MB
# Or download all models
make download-models

Usage:

# Transcribe a file
swift run argmax-cli transcribe --model-path "Models/whisperkit-coreml/openai_whisper-large-v3-v20240930_626MB" --audio-path "path/to/your/audio.{wav,mp3,m4a,flac}"

# Stream from microphone
swift run argmax-cli transcribe --model-path "Models/whisperkit-coreml/openai_whisper-large-v3-v20240930_626MB" --stream

You can install the command line application for WhisperKit using Homebrew.

brew install whisperkit-cli

Before using fastlane, ensure you have the latest version of the Xcode command line tools installed on your system.

xcode-select --install

TTSKit is a Core ML-based framework for real-time text-to-speech on Apple silicon. It supports macOS 15.0+ and iOS 18.0+.

Basic Usage: Initialize TTSKit() (which automatically downloads the 0.6B model on first run) and call generate(text:).

Model Variants:

• .qwen3TTS_0_6b: Fast, runs on all platforms (~1 GB).
• .qwen3TTS_1_7b: Higher quality, macOS only (~2.2 GB), supports natural language style instructions.

import TTSKit

Task {
    let tts = try await TTSKit()
    let result = try await tts.generate(text: "Hello from TTSKit!")
    print("Generated \(result.audioDuration)s of audio at \(result.sampleRate)Hz")
}

The Argmax CLI includes a local server that implements the OpenAI Audio API for transcription and translation, supporting output streaming.

Build the server:

make build-local-server
# OR
BUILD_ALL=1 swift build --product argmax-cli

Start the server:

# Default settings
BUILD_ALL=1 swift run argmax-cli serve

# Custom host, port, and model
BUILD_ALL=1 swift run argmax-cli serve --host 0.0.0.0 --port 8080 --model tiny --verbose

API Endpoints:

• POST /v1/audio/transcriptions: Transcribe audio to text.
• POST /v1/audio/translations: Translate audio to English.

Benchmarks are managed via a Makefile and executed using Fastlane. You can run different configurations depending on your needs:

• Debug Mode: Run with DEBUG=true to check for potential errors.
• Full Mode: Run the standard full benchmark suite.
• Target Specific Devices: Use the DEVICES option with a comma-separated list of device names (obtained from make list-devices).

Note: An active developer account and Developer Mode enabled on physical devices are required.

To run the WhisperKit benchmark test suite, clone the repository using the following command:

git clone git@github.com:argmaxinc/argmax-oss-swift.git