--- language: - en - de license: other license_name: nyra-health-non-commercial-research license_link: LICENSE.md pipeline_tag: automatic-speech-recognition library_name: crisperwhisper tags: - speech-recognition - verbatim - disfluency - whisper - ctranslate2 - word-timestamps - cpp - ggml - c++ - cuda - cpu - arxiv:2607.18934 --- ![crisperwhispercpp](https://cdn-uploads.huggingface.co/production/uploads/63473b59e5c0717e6737b872/ptbxcjvQ--WxdCImZUIvt.png) # CrisperWhisper 2.0 GGML Unofficial FP16 GGML conversions of the public CrisperWhisper 2.0 models for [CrisperWhisper.cpp](https://github.com/Saganaki22/CrisperWhisper.cpp). CrisperWhisper.cpp provides native C++ inference on Windows and Linux with CPU and NVIDIA CUDA backends. Python is not required at runtime. > This repository is an unofficial community conversion. It is not affiliated > with, endorsed by, or maintained by Nyra Health or Nyra Labs. ## Model files | Model | GGML file | Approximate size | Original model | | --- | --- | ---: | --- | | Small | [`ggml-crisperwhisper-small-f16.bin`](https://huggingface.co/drbaph/CrisperWhisper2.0-GGML/resolve/main/ggml-crisperwhisper-small-f16.bin?download=true) | 488 MB | [`nyralabs/CrisperWhisper2.0_small`](https://huggingface.co/nyralabs/CrisperWhisper2.0_small) | | Medium | [`ggml-crisperwhisper-medium-f16.bin`](https://huggingface.co/drbaph/CrisperWhisper2.0-GGML/resolve/main/ggml-crisperwhisper-medium-f16.bin?download=true) | 1.53 GB | [`nyralabs/CrisperWhisper2.0_medium`](https://huggingface.co/nyralabs/CrisperWhisper2.0_medium) | | Turbo | [`ggml-crisperwhisper-turbo-f16.bin`](https://huggingface.co/drbaph/CrisperWhisper2.0-GGML/resolve/main/ggml-crisperwhisper-turbo-f16.bin?download=true) | 1.62 GB | [`nyralabs/CrisperWhisper2.0_turbo`](https://huggingface.co/nyralabs/CrisperWhisper2.0_turbo) | | Large | [`ggml-crisperwhisper-large-f16.bin`](https://huggingface.co/drbaph/CrisperWhisper2.0-GGML/resolve/main/ggml-crisperwhisper-large-f16.bin?download=true) | 3.09 GB | [`nyralabs/CrisperWhisper2.0_large`](https://huggingface.co/nyralabs/CrisperWhisper2.0_large) | These are FP16 conversions, not integer-quantized models. The gated [`CrisperWhisper2.0_large_pro`](https://huggingface.co/nyralabs/CrisperWhisper2.0_large_pro) checkpoint is not included. Access to Large Pro requires approval and is subject to its separate commercial model terms. ## Native C++ runtime Download the native runtime from: - [CrisperWhisper.cpp on GitHub](https://github.com/Saganaki22/CrisperWhisper.cpp) - [CrisperWhisper.cpp v1.1.0 release](https://github.com/Saganaki22/CrisperWhisper.cpp/releases/tag/v1.1.0) Available v1.1.0 packages: - Windows x64 CUDA for RTX 30, RTX 40, and RTX 50 GPUs - Windows x64 portable CPU - Windows x64 AVX2 CPU - Linux x64 portable CPU - Linux x64 AVX2 CPU Linux CUDA is available as a source build. Model files are downloaded separately and are not duplicated inside the GitHub release archives. ## Download ```bash hf download drbaph/CrisperWhisper2.0-GGML \ ggml-crisperwhisper-small-f16.bin \ --local-dir models ``` Replace `small` with `medium`, `turbo`, or `large` to download another model. ## Windows usage ```powershell .\crisper-whisper.exe ` --model .\models\ggml-crisperwhisper-small-f16.bin ` --file .\audio.wav ` --mode verbatim ` --language en ``` Optional supervised word timestamps and JSON: ```powershell .\crisper-whisper.exe ` --model .\models\ggml-crisperwhisper-small-f16.bin ` --file .\audio.wav ` --mode verbatim ` --language en ` --word-timestamps ` --json ``` ## Linux usage ```bash ./crisper-whisper \ --model ./models/ggml-crisperwhisper-small-f16.bin \ --file ./audio.wav \ --mode verbatim \ --language en ``` Optional supervised word timestamps and JSON: ```bash ./crisper-whisper \ --model ./models/ggml-crisperwhisper-small-f16.bin \ --file ./audio.wav \ --mode verbatim \ --language en \ --word-timestamps \ --json ``` ## Audio input The native CLI accepts WAV, MP3, FLAC, and Ogg Vorbis. Audio is decoded, downmixed to mono, and resampled to 16 kHz automatically in memory. FFmpeg and a separate preprocessing script are not required. ## Transcription modes Verbatim transcription preserves spoken disfluencies: ```bash crisper-whisper \ -m models/ggml-crisperwhisper-small-f16.bin \ -f audio.wav \ --mode verbatim ``` Intended transcription produces a cleaner intended transcript: ```bash crisper-whisper \ -m models/ggml-crisperwhisper-small-f16.bin \ -f audio.wav \ --mode intended ``` ## Word timestamps Supervised cross-attention word timestamps are optional: ```bash crisper-whisper \ -m models/ggml-crisperwhisper-small-f16.bin \ -f audio.wav \ --word-timestamps \ --json ``` Without `--word-timestamps`, no alignment pass or additional timestamp model context is created. Normal transcription keeps GGML Flash Attention enabled. Timestamp alignment uses an additional teacher-forced pass without Flash Attention because the fused Flash Attention path does not expose the decoder-to-audio attention matrix required for supervised alignment. Long-form timestamp output includes seam ownership, duplicate-prefix removal, and monotonic timestamp correction to prevent duplicated words, overlapping word chunks, and invalid `start > end` timestamps at chunk boundaries. ## Original project and paper - [Official CrisperWhisper repository](https://github.com/nyrahealth/CrisperWhisper) - [Official Small model](https://huggingface.co/nyralabs/CrisperWhisper2.0_small) - [Official Medium model](https://huggingface.co/nyralabs/CrisperWhisper2.0_medium) - [Official Turbo model](https://huggingface.co/nyralabs/CrisperWhisper2.0_turbo) - [Official Large model](https://huggingface.co/nyralabs/CrisperWhisper2.0_large) - [CrisperWhisper paper](https://arxiv.org/abs/2607.18934) - [Native C++ implementation](https://github.com/Saganaki22/CrisperWhisper.cpp) - [whisper.cpp](https://github.com/ggml-org/whisper.cpp) ## Limitations - These files are conversions of the original checkpoints, not separately trained models. - Small, medium, turbo, and large have different speed, memory, and accuracy characteristics. - FP16 files require more storage and memory than integer-quantized models. - Output can differ slightly from Python/Transformers because of backend and numerical implementation differences. - Speculative draft-model decoding is not currently implemented. - Large Pro is not included. ## License These conversions do not change or replace the original model license. The Small, Medium, Turbo, and Large model weights in this repository remain governed by the [Nyra Health Non-Commercial Research License](https://huggingface.co/nyralabs/CrisperWhisper2.0_large/blob/main/LICENSE.md). Review that license before downloading, using, modifying, or redistributing the converted model files. These models are intended for permitted non-commercial research use. The original models, architecture, research, and associated intellectual property belong to their respective authors and rights holders.