Whisper.spm
https://github.com/ggerganov/whisper.spm- Category
- AI Tools
- Rank
- No. 1677Tools index
Previous survey · No. 1685 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- ggerganov
- GitHub
- 191 stars
- Date
About
Drop whisper.cpp speech-to-text into iOS or macOS apps as a Swift Package Manager dependency, no manual build steps.
What it does
Whisper.spm packages a native speech recognition engine as a library that Swift and Objective-C applications can import. Applications load a model, supply floating-point audio samples, configure transcription, and read text segments.
Why it's ranked here
The package offers a direct native path from Apple application code to local transcription, with example clients and automated builds. Its value is now limited by an explicit maintenance warning, a branch-based installation requirement, and disabled Metal support.
What's good
The library exposes the same public C interface to Swift and Objective-C. It enables Accelerate and Core ML, allows Core ML fallback, includes runnable examples for both languages, and checks both targets in macOS automation.
Tradeoffs
The repository says it will be archived and directs users to another package. Installation must track the master branch because other dependency rules trigger unsafe-build-flag errors. Metal sources are excluded, despite the package targeting Apple platforms.
How to use it well
Use it for an existing Apple application that needs native, local transcription and can tolerate a maintenance migration. Pin your integration workflow carefully and validate both language targets. Bring your own model files and audio capture pipeline, since the examples assume both inputs already exist.
Technical notes+
Package.swift defines a Swift tools 5.3 library product named whisper, compiles C and C++ sources under Sources/whisper, publishes headers from Sources/whisper/include, links Accelerate, enables GGML_USE_ACCELERATE and Core ML flags, and excludes Sources/whisper/ggml-metal.m plus Sources/whisper/ggml-metal.metal. Sources/test-swift/main.swift and Sources/test-objc/main.m exercise model loading, PCM transcription, segment retrieval, timing output, and cleanup through the C API. .github/workflows/swift.yml builds on macOS and runs both executables. README.md requires the master branch because of unsafe build flags and announces planned archival.
Observed
- License
- MIT License
- Primary language
- Native implementation in C and C++, with Swift and Objective-C example clients
- Packaging
- Swift Package Manager library product named whisper
- Interface
- Native library with public C headers, importable from Swift and Objective-C
- Platform support
- On Arm builds, declares macOS 11, iOS 14, watchOS 4, and tvOS 14
- Build structure
- Defines separate Swift and Objective-C test targets and runs both in macOS CI
- Apple integration
- Links Accelerate and enables Core ML with fallback; Metal sources are excluded
Read from README.md, Makefile, Package.swift, Sources/whisper/whisper.cpp, Sources/whisper/ggml-alloc.c, Sources/test-swift/main.swift, Sources/whisper/ggml-backend.c, LICENSE, Makefile-tmpl, Sources/test-objc/main.m, .github/workflows/swift.yml, Sources/whisper/ggml-impl.h, Sources/whisper/ggml-alloc.h, Sources/whisper/ggml-metal.h.
What it can do
Convert speech audio to text
Audio file or stream → Transcribed text
Integrate speech-to-text functionality into iOS apps
Swift Package Manager dependency → iOS app with speech recognition capabilities
Integrate speech-to-text functionality into macOS apps
Swift Package Manager dependency → macOS app with speech recognition capabilities
Add whisper.cpp engine to Swift projects
Swift Package Manager configuration → Project with embedded whisper.cpp functionality
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.