faster-whisper

faster-whisper is a reimplementation of OpenAI's Whisper automatic speech recognition model using CTranslate2, a fast inference engine for Transformer models. It loads the same Whisper weights (tiny through large-v3, plus distil-large-v3) but runs them through optimized CTranslate2 kernels with optional 8-bit quantization, delivering the same transcription accuracy with substantially lower latency and memory use. It is the standard pragmatic choice for Python transcription services, subtitle pipelines, and batch audio processing, and forms the transcription core of tools like WhisperX. Maintained by SYSTRAN, the company behind CTranslate2 and OpenNMT.

Install from PyPI (Python 3.9+):

```bash

pip install faster-whisper

```

GPU (NVIDIA): the latest ctranslate2 releases support only CUDA 12 with cuDNN 9. Install the NVIDIA libraries first, e.g. pip install nvidia-cublas-cu12 nvidia-cudnn-cu12==9.*, or pin ctranslate2==3.24.0 for CUDA 11 setups.

Basic usage:

```python

from faster_whisper import WhisperModel

model = WhisperModel("large-v3", device="cuda", compute_type="float16")

# or CPU with 8-bit quantization:

# model = WhisperModel("large-v3", device="cpu", compute_type="int8")

segments, info = model.transcribe("audio.mp3", beam_size=5)

print("Detected language '%s'" % info.language)

for segment in segments:

print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

```

Note: segments is a generator — transcription runs as you iterate. Model weights download automatically from Hugging Face on first use.