CrispASR

Why we included this project

CrispASR is a single C++ binary built on ggml that handles both speech-to-text and text-to-speech, so you get one tool for the two directions of speech. It runs a wide range of open-weights ASR models, including Whisper, NVIDIA Parakeet and Canary, Mistral Voxtral, and Cohere Transcribe, plus dozens of TTS engines. The backend is picked at the command line or auto-detected from the GGUF file, and `-m auto` downloads weights on first use, which removes most of the setup friction. There is no Python or PyTorch dependency, which makes it attractive for teams that want to embed transcription in an existing service; the HTTP server mode, C-ABI, and bindings for Python, Rust, Go, and other languages cover that path, and a WebAssembly build runs everything client-side in a browser. If you would rather have one consistent interface across many speech models than juggle separate per-model tools, this is a practical fit.