babelscribe

babelscribe — free, offline speech-to-text and subtitles on any GPU (AMD, NVIDIA, Intel, Apple)

babelscribe turns any audio or video file into subtitles (SRT, VTT) and text, in any of Whisper's 99 languages, on your own graphics card — AMD Radeon, NVIDIA, Intel Arc or Apple Silicon — or the CPU. No CUDA needed, no cloud, free (MIT). It runs OpenAI Whisper through whisper.cpp with Vulkan, CUDA or Metal, and works from the command line, from Python, or from AI apps (Claude, Codex, Antigravity, Gemini CLI, Cursor) as an MCP server.

pip install babelscribe          # Python 3.9+

babelscribe interview.mp4 # auto-detects language, picks your best GPU -> interview.srt + interview.json

babelscribe is a thin, friendly layer on top of whisper.cpp. It exists because getting fast Whisper on a non-NVIDIA card (e.g. an AMD Radeon on Windows) still means building whisper.cpp with the Vulkan SDK yourself. babelscribe downloads a prebuilt whisper-cli for your system and handles the rest.

A 19-minute English TED talk transcribes in 48 seconds on an AMD Radeon RX 9070 XT with 1.5% word error — see the benchmark below.

Use it from an AI app (MCP) — no terminal needed

babelscribe is also an MCP server, so any AI app that supports local MCP servers can transcribe files on your computer with your GPU. Ask in plain words — *"make Thai subtitles for the newest video in my Downloads"* — and the app finds the file, transcribes it, writes .srt / .txt next to it, and can then proofread names, translate or summarise the text for you.

Tools: transcribe (file → subtitles + text), find_media (newest audio/video in Downloads, Videos, Desktop, …), list_devices, list_languages_and_models.

Claude Desktop (one click): download babelscribe.mcpb from the latest release and double-click it (or *Settings → Extensions → Install extension*).

Everything else runs the same command — uv fetches babelscribe for you: uvx babelscribe mcp

{

"mcpServers": { "babelscribe": { "command": "uvx", "args": ["babelscribe", "mcp"], "timeout": 3600000 } } }

(`timeout` is in milliseconds and only some apps read it.) Already installed with pip? Use `"command": "babelscribe", "args": ["mcp"]`.

Web-only chat apps (e.g. grok.com, chatgpt.com) can only reach servers on the internet, not your computer, so they can't use your GPU or files.

Benchmark: real talks, human captions as the answer key

Four TED / TEDx talks, scored against the human-made captions in the spoken language (bench/bench.py, reproducible). Error = word error rate (WER) for space-separated languages, character error rate (CER) for Japanese and Thai. GPU: AMD Radeon RX 9070 XT via Vulkan. *default* = large-v3-turbo; *accurate* = --accurate (see the FLEURS section).

How to read it: TED captions are edited for reading (fillers dropped, light rewording), so these numbers are an upper bound — most of the Spanish "errors" are the speaker's actual words versus the tidied caption. Thai is genuinely harder: fast, casual speech with slang; --accurate (Pathumma Whisper text + turbo timing) cuts its error by more than a quarter. Talks are used only to measure accuracy; their transcripts are not redistributed (TED content is CC BY-NC-ND).

Benchmark: 16 languages on FLEURS

Google FLEURS test set, 50 utterances per language, human-verified verbatim transcripts (bench/fleurs.py, reproducible). Same normaliser for every language (Whisper's rule: lower-case, drop punctuation and non-spacing marks, numbers spelled out). WER for space-separated languages, CER for ja / zh / ko / th.

\* Some FLEURS clips contain more speech than their reference transcript, so a correct model is charged for words the reference leaves out. *Edge-trimmed* ignores extra words before the first / after the last reference word; both numbers are stored by bench/fleurs.py. 50 utterances per language means differences under ~0.5 points are noise.

Also measured and not used, because large-v3 was as good or better: large-v2 (all languages), PhoWhisper-large (vi 19.0%), whisper-large-v3 dialectal / code-switching Arabic fine-tunes (12.2% / 15.2%), Typhoon Whisper (th 11.9%), Thonburian Whisper (th 9.1% — kept as an option), Vaani Hindi (16.5%).

Why babelscribe (vs. what already exists)

babelscribe does not replace those projects — it stands on whisper.cpp and simply removes the hard parts: compiling for your GPU, converting media, picking the right device, avoiding the long-file repeat bug, and combining a language-specific fine-tune with accurate timestamps.

FAQ

How do I run Whisper on an AMD GPU on Windows? pip install babelscribe, then babelscribe video.mp4. It downloads a Vulkan build of whisper.cpp that runs on AMD Radeon (and Intel Arc / NVIDIA) cards on Windows and Linux — no ROCm, no CUDA, no compiling.

How do I make subtitles (SRT) from a video for free, offline? babelscribe video.mp4 -f srt writes video.srt next to the video. Nothing is uploaded; it runs on your own computer.

What is the most accurate free transcription for Thai? babelscribe video.mp4 -l th --accurate — Pathumma Whisper (NECTEC) for the text plus Whisper turbo for timing: CER 8.9% on Google FLEURS vs 15.9% for plain Whisper turbo. Thonburian Whisper is available too (--text-model thai-thonburian).

Can Claude / ChatGPT Codex / Gemini transcribe a video on my computer? Yes — add babelscribe as an MCP server (see *Use it from an AI app*). Claude Desktop installs it with one click from babelscribe.mcpb. The AI app can then find a file, transcribe it on your GPU, and proofread, translate or summarise the text.

How fast is it? About 20x real time with Whisper large-v3-turbo on an AMD Radeon RX 9070 XT: a 19-minute talk in 48 seconds.

Which languages are supported? All 99 Whisper languages, with automatic language detection. FLEURS error rates for 16 of them are in the benchmark table.

Is it better than faster-whisper or WhisperX? Those are excellent on NVIDIA GPUs; on AMD / Intel GPUs they fall back to the CPU. babelscribe's niche is any GPU, zero setup, and better Thai / Hindi through community fine-tunes. If you have an NVIDIA card and like Python, faster-whisper is a fine choice.

Languages

All 99 languages Whisper was trained on, auto-detected or forced with -l: af am ar as az ba be bg bn bo br bs ca cs cy da de el en es et eu fa fi fo fr gl gu ha haw he hi hr ht hu hy id is it ja jw ka kk km kn ko la lb ln lo lt lv mg mi mk ml mn mr ms mt my ne nl nn no oc pa pl ps pt ro ru sa sd si sk sl sn so sq sr su sv sw ta te tg th tk tl tr tt uk ur uz vi yi yo yue zh

Accuracy follows Whisper's own training data: excellent for high-resource languages (English, Spanish, Japanese, …), weaker for low-resource ones. That is what hybrid mode is for — a community fine-tune for one language can be plugged in with one line in babelscribe/models.py (Thai and Hindi so far). PRs adding fine-tunes for other languages are the most valuable contribution.

Limitations (honest)

• Tested end to end so far on an AMD Radeon RX 9070 XT (Windows, Vulkan). CUDA, Linux and macOS builds are produced by CI; reports from those machines are welcome.
• Hybrid mode runs two models, so it is slower (≈1 min per minute of audio with a large fine-tune on that GPU).
• Proper nouns can still be misspelled — check names before publishing subtitles.

Features

• Any input — mp4, mkv, mov, mp3, wav, m4a… (ffmpeg is bundled through imageio-ffmpeg).
• Any language — -l auto detects it; or pass `-l