This public link is valid for 7 days and shares a thread, including any personal information you added. This link or copies made by others cannot be deleted. If you share with third parties, their policies apply. Can’t copy the link right now. Try again later. ggerganov/whisper.cpp at main - Hugging Face
Alternatively, if you have cloned the repository, use the included shell script: sh ./models/download-ggml-model.sh medium Use code with caution. 2. Run the Model
Look for whisper-medium-gguf.bin or simply download the medium model via whisper.cpp ’s built-in script:
-osrt : Output the transcription directly into a SubRip ( .srt ) subtitle file, perfect for video editing. ggml-medium.bin
Only if you no longer need the AI model. Without this file, the inference program won’t work. If you downloaded it manually, you can always re‑download it later.
The "medium" model is often considered the "sweet spot" for users who need higher accuracy than the "base" or "small" models but cannot afford the massive hardware requirements of the "large" models.
The Whisper model was originally released by OpenAI as a massive, resource-hungry PyTorch file. To make it run on everyday hardware like laptops and phones, developers created the . This specialized format allows the model to run efficiently in C++, enabling users to transcribe audio offline without sending data to the cloud . 2. The Quest for Balance This public link is valid for 7 days
| Model | VRAM/RAM | Speed (Real-time factor) | WER (Word Error Rate) | Use case | |-------|----------|--------------------------|----------------------|-----------| | tiny | ~150 MB | 0.10x (10x faster) | ~25% (poor) | Voice commands, real-time keyword spotting | | base | ~300 MB | 0.15x | ~15% | Simple dictation, low-resource devices | | small | ~500 MB | 0.25x | ~8% | General transcription, podcasts | | | ~700 MB | 0.50x (2x real-time) | ~5% | Legal/medical drafts, multilingual meetings | | large | ~1.5 GB | 1.0x (real-time) | ~3% (best) | High-stakes transcription, research |
Expect to need at least 4GB of free RAM to run ggml-medium.bin comfortably, although 8GB+ is recommended for optimal performance, especially if using CPU-only mode.
When working with whisper.cpp , you have several size options: Tiny, Base, Small, Medium, and Large. While ggml-large-v3.bin is the most accurate, it is often overkill for daily use. Can’t copy the link right now
Running ggml-medium.bin requires more resources than smaller models, but it does not demand a dedicated server.
: Highly accurate but massive (often over 3GB), requiring heavy GPU power and significant memory.
While whisper-tiny is incredibly fast, it struggles with accents, technical jargon, and background noise. Conversely, whisper-large is highly accurate but painfully slow on non-enterprise hardware. ggml-medium.bin sits perfectly in the middle, offering professional-grade transcription accuracy with swift processing times. 2. Complete Local Privacy