Basic Transcription Example

A Swift command-line application that demonstrates how to use Moonshine Voice for transcription, mirroring the functionality of the Python basic_transcription.py script.

Features

• Transcribe audio files offline without streaming
• Transcribe audio files with streaming (event-based)
• Support for multiple languages and model architectures
• Command-line argument parsing

Building

Using Swift Package Manager

cd examples/macos/MicTranscription
swift build

Using Xcode

An Xcode project is available for building and running from the Xcode IDE:

1. Open MicTranscription.xcodeproj in Xcode
2. Select the MicTranscription scheme
3. Build and run (⌘R) or use the Product menu

The Xcode project is generated from project.yml using xcodegen. To regenerate it:

cd examples/macos/MicTranscription
xcodegen generate

Running

# Use default test file (test-assets/two_cities.wav)
swift run MicTranscription

# Specify input files
swift run MicTranscription path/to/audio1.wav path/to/audio2.wav

# Specify language
swift run MicTranscription --language en path/to/audio.wav

# Specify model architecture
swift run MicTranscription --model-arch tiny path/to/audio.wav

# Show help
swift run MicTranscription --help

Command-Line Options

• --language, -l LANGUAGE: Language to use for transcription (default: en)
• --model-arch, -m ARCH: Model architecture: tiny, base, tiny-streaming, base-streaming, small-streaming, medium-streaming
• --help, -h: Show help message

Supported Languages

• en / english - English
• ja / japanese - Japanese
• es / spanish - Spanish
• ar / arabic - Arabic
• ko / korean - Korean
• vi / vietnamese - Vietnamese
• uk / ukrainian - Ukrainian
• zh / chinese - Chinese

Example Output

The application will:

1. Transcribe each input file offline (non-streaming) and print the results
2. Transcribe each input file with streaming, showing events as they occur:
   • Line started events
   • Line text changed events
   • Line completed events

Notes

• The application expects model files to be in test-assets/{model-name}/ directory
• WAV files must be PCM format (16-bit, 24-bit, or 32-bit)
• Multi-channel audio is automatically mixed down to mono