A faster-whisper alternative that uses your AMD or Intel GPU
Short answer: faster-whisper is excellent on NVIDIA. If your GPU is AMD Radeon, Intel Arc or Apple Silicon, pip install babelscribe gives you GPU-speed Whisper through whisper.cpp (Vulkan / Metal) with no CUDA.
Comparison
| babelscribe | faster-whisper / WhisperX | |
|---|---|---|
| GPU support | Vulkan (AMD, Intel, NVIDIA), CUDA, Metal (Apple), CPU | NVIDIA CUDA; otherwise CPU |
| Windows setup | pip install babelscribe; prebuilt whisper-cli downloaded for you | pip, plus CUDA / cuDNN for GPU |
| Video in, subtitles out | Yes, ffmpeg bundled; SRT, VTT, TXT, JSON | Yes (library; you write the glue) |
| Better Thai / Hindi text | Yes, --accurate (hybrid fine-tune + turbo timing) | Manual |
| AI-app integration | MCP server for Claude, Codex, Gemini CLI, Cursor | Third-party wrappers |
If you have an NVIDIA card and like Python APIs, faster-whisper is a fine choice. babelscribe does not replace it; it covers the GPUs it does not. Speed: about 20x real time on an AMD Radeon RX 9070 XT (19-minute talk in 48 s).
pip install babelscribe
babelscribe interview.mp4 -f srtFAQ
Is there a faster-whisper alternative for AMD GPUs?
Yes. babelscribe runs Whisper through whisper.cpp with Vulkan, which works on AMD Radeon and Intel Arc GPUs.
Does faster-whisper work on AMD?
Its GPU path targets NVIDIA CUDA; on AMD it runs on the CPU. babelscribe uses the AMD GPU through Vulkan.
Is babelscribe as accurate as faster-whisper?
Both run the same Whisper models, so base accuracy is comparable. babelscribe adds an --accurate mode with language-specific fine-tunes for Thai and Hindi.
babelscribe is free and open source (MIT) · Privacy