Whisper Large v3 · OpenASR
OpenAI's most accurate Whisper, the v3 large checkpoint
Native speech-to-text in the OpenASR runtime — engineered for peak performance on CPU & GPU, no Python at inference time.
✨ Highlights
- 🎧 Multilingual ASR — transcribes a wide range of languages and can translate speech to English
- 🏆 1.55B parameters — the full-size Whisper, OpenAI's highest-accuracy checkpoint
- 🔁 v3 improvements — trained on a larger, more diverse corpus with 128 mel bins for better robustness
- 🦀 Native in OpenASR —
.oasrpacks run with no Python at inference, engineered for peak performance on CPU & GPU
🚀 Quickstart
# 1. Install the OpenASR CLI · https://openasr.org
# 2. Pull a build (pick a quant — see the table below)
openasr pull whisper-large-v3:q8
# 3. Transcribe
openasr transcribe audio.wav --model whisper-large-v3
All builds for this model:
openasr pull whisper-large-v3:fp16
openasr pull whisper-large-v3:q8
openasr pull whisper-large-v3:q4
📦 Available builds
| Quant | File (.oasr) | Size | RAM peak | RTF · M1 CPU | RTF · M1 GPU | JFK ΔWER vs fp16 |
|---|---|---|---|---|---|---|
| fp16 | whisper-large-v3-fp16.oasr | 3.09 GB | 4.70 GB | 1.17× | 1.13× | 0.0% |
| q8_0 | whisper-large-v3-q8_0.oasr | 1.71 GB | 4.05 GB | 0.65× | 0.46× | 0.0% |
| q4_k | whisper-large-v3-q4_k.oasr | 1.29 GB | 2.78 GB | n/a | n/a | n/a |
RTF = real-time factor on the fixed 11s JFK clip (lower is faster); RAM peak measured per pack in an isolated subprocess. JFK ΔWER compares each quantized build's JFK transcript to this model's fp16 JFK transcript, so it measures quantization drift rather than absolute recognition accuracy. q8_0 is the recommended default — near-reference quality at a fraction of the footprint.
🧠 About Whisper Large v3
Whisper Large v3 is OpenAI's 1.55B-parameter multilingual Whisper checkpoint, the most accurate
member of the family. It uses the standard Whisper encoder-decoder architecture for automatic
speech recognition and speech translation; v3 was trained on a larger and more diverse labelled
corpus and uses 128 mel-frequency bins, improving robustness across languages and conditions
over earlier large checkpoints. This OpenASR repo repackages the original
openai/whisper-large-v3 weights as .oasr packs that run natively in the OpenASR runtime with
no Python at inference time. For most users the q8_0 build is the recommended default; q4_k is
for tighter memory budgets and fp16 is for verification or maximum fidelity. For a faster
large-grade option, see the distilled whisper-large-v3-turbo.
⚙️ How these packs were made
Converted from openai/whisper-large-v3 with the OpenASR importer:
openasr model-pack import whisper <src> <out>.oasr \
--package-id whisper-large-v3 --quantization {fp16,q8-0,q4-k}
The .oasr container is GGUF-backed; packs use zero-copy mmap weight binding and graph
buffer reuse to keep peak memory low.
⚖️ License
These packs inherit the upstream model's license: Apache-2.0 (source). OpenASR packaging retains the upstream copyright and NOTICE; the only modifications are format conversion and quantization.
🙏 Acknowledgements
This pack is a redistribution of Whisper Large v3, released by OpenAI
(openai/whisper-large-v3).
All credit for the original model, training recipe, and weights belongs to OpenAI. The
upstream Hugging Face model card declares Apache-2.0 licensing; OpenASR only converts the
weights into .oasr packages and adds quantized builds for local runtime use.
🔗 Links
- 🦀 OpenASR — https://github.com/QuintinShaw/openasr
- 🌐 Website — https://openasr.org
- 🤗 Upstream model — openai/whisper-large-v3