CrispStrobe / CrispASR
One C++ binary running 119 ASR/TTS backends on ggml with zero Python deps, WebAssembly support, and bindings for six languages.
星标趋势
AI 分析
项目摘要
CrispASR is a C++ ggml runtime hub that unifies 119 speech backends (62 TTS engines) into a single binary for multilingual ASR, TTS, and translation. Forked from whisper.cpp, it eliminates Python dependencies while supporting WebAssembly, multiple language bindings, and auto-detection of GGUF models.
为什么值得关注
It consolidates a vast array of speech architectures (Whisper, Parakeet, Voxtral, Canary, Kokoro, Qwen3, and more) into one lightweight C++ binary with zero Python dependencies, offering WebAssembly deployment and bindings for Python, Rust, Dart, Go, Ruby, and Java — making offline speech AI accessible across virtually any platform.
优势
- Single binary with 119 backends (62 TTS) and zero Python dependencies
- WebAssembly compilation for browser-based client-side inference
- Broad language bindings: Python, Rust, Dart, Go, Ruby, Java, plus C-ABI and HTTP server
- Auto-detection of model architecture from GGUF files
- Highly active development with 63 releases in 6 months and CI/test coverage
局限性
- Fork of whisper.cpp — derivative rather than ground-up innovation
- No Docker support for containerized deployment
- Relatively new project (created March 2026) with limited long-term track record
使用场景
- Offline speech recognition in resource-constrained or air-gapped environments
- Browser-based transcription and TTS via WebAssembly
- Cross-platform desktop and mobile apps through the CrisperWeaver Flutter ecosystem
- Embedded systems requiring lightweight C++ speech inference without Python overhead
- Multilingual ASR and TTS pipelines with unified CLI and API