Practical overview
When it fits
Use Speech Swift when building or running speech workflows on supported Apple Silicon hardware and you want model inference to run locally rather than through a cloud API. Choose the relevant package or command-line path for the task.
Who it is for
- Developers adding speech recognition or synthesis to macOS and iOS apps.
- Swift developers building streaming dictation interfaces with live partial and final transcription.
- Apple Silicon users who want to run documented speech tasks through the project’s command-line tools.
Documented uses
- Transcribe audio in a Swift application or from the command line.
- Display live transcription partials and final results in a SwiftUI interface.
- Add text-to-speech or speech-to-speech capabilities to a local workflow.
- Use voice activity detection or speaker diarization as part of a speech application.
Requirements & limits
The repository is public and identifies the project as Apache-2.0 licensed. Documented prerequisites include Swift 6+, Xcode 16+ with the Metal Toolchain, macOS 15+ or iOS 18+, and Apple Silicon (M1/M2/M3/M4).
- It is a developer toolkit rather than a finished end-user transcription app.
- Model weights download from Hugging Face on first use and are cached; offline mode is documented for models already cached.
- The repository offers separate model package products, so the application can import selected capabilities.
- The documented Homebrew route requires native ARM Homebrew at /opt/homebrew; Rosetta/x86_64 Homebrew is unsupported.
- A source build also compiles the MLX Metal shader library, which the project says is required for GPU inference.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Add streaming transcription to a Swift app
Use the documented Parakeet streaming model and optional SpeechUI components to display partial and final transcription results.
Before you start
- Swift 6 or later and Xcode 16 or later with the Metal Toolchain.
- macOS 15 or later or iOS 18 or later on Apple Silicon (M1, M2, M3 or M4).
- A Swift Package Manager project and audio samples supplied to the transcription stream.
Steps
Add the speech-swift package from its GitHub URL to the project’s Swift Package Manager dependencies.
Include the ParakeetStreamingASR product; add SpeechUI if using its documented transcription view and store.
Import ParakeetStreamingASR and create a model with ParakeetStreamingASRModel.fromPretrained().
Pass audio samples and their sample rate to transcribeStream(audio:sampleRate:), then handle each partial’s text and isFinal value.
Expected outcome
The app can receive streaming transcription updates, including partial text and final results; SpeechUI provides a view and adapter store for presenting those updates.
Guide limitations
- Model weights download on first use and are cached locally; offline loading is available only when the weights are already cached.
- The documented audio visualization and playback role is handled with AVFoundation, not SpeechUI.
- This guide describes the documented integration pattern and is not a compatibility or execution test.
Install the speech command-line tools with Homebrew
Set up the documented Speech Swift command-line package for tasks such as transcribing an audio recording.
Before you start
- Native ARM Homebrew installed at /opt/homebrew; Rosetta/x86_64 Homebrew is not supported.
- Apple Silicon device and the project’s stated macOS, Swift and Xcode requirements.
Steps
Install the package with brew install speech.
Use a documented command such as speech transcribe recording.wav to request transcription.
For a different documented transcription engine, select one with the --engine option, for example speech transcribe recording.wav --engine cohere.
Expected outcome
The speech command-line tool is installed, and the documented commands show how to request transcription with the default or a selected engine.
Guide limitations
- The README does not specify model download sizes or processing performance for these commands.
- The Homebrew instructions require native ARM Homebrew and do not support Rosetta/x86_64 Homebrew.
- This guide describes repository instructions and is not a successful installation test.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- soniqo/speech-swift · Official homepage title · retained in its source language