faster-whisper
Faster Whisper speech-to-text transcription using CTranslate2, a Python reimplementation of OpenAI's Whisper.
- Stars
- 26.4k
- Forks
- 2.2k
- Licence
- MIT
- Language
- Python
- Last update
- 5 days ago
- Type
- Library
What it is
faster-whisper is a Python library that transcribes speech in audio files to text, using OpenAI's Whisper models. It reimplements Whisper on top of CTranslate2, an inference engine for Transformer models. The README says it is up to 4 times faster than openai/whisper for the same accuracy and uses less memory, with further speedups from 8-bit quantization.
- Transcribe an audio file to text with timestamps per segment
- Generate word-level timestamps for subtitles or search indexes
- Skip silent sections of audio before transcribing using the built-in Silero VAD filter
- Run batched transcription on a GPU with BatchedInferencePipeline for higher throughput
Great for
- Python developers who need to add speech-to-text to an application
- People transcribing large batches of audio who want lower time or memory use
- Users with an NVIDIA GPU or a CPU who want to run Whisper locally
- Anyone who wants word-level timestamps or voice activity filtering on transcripts
Not ideal for
- People who want a ready-made transcription app with a graphical interface, since this is a code library (the README shows only Python usage)
- Users with an older NVIDIA setup, because the README says the latest ctranslate2 supports only CUDA 12 and cuDNN 9, and older setups need pinned versions
- Users who need Python 3.8 or older, because the README requires Python 3.9 or greater
How to get it
- 1
Install with a package manager
You'll need: Python 3.9 or greater · NVIDIA cuBLAS for CUDA 12 and cuDNN 9 (only for GPU use)
Install from PyPI
pip install faster-whisper - 2
Run it
- Install the package with pip install faster-whisper
- Create a WhisperModel with a model size such as large-v3, choosing device and compute_type
- Call model.transcribe("audio.mp3") to get segments and info
- Iterate over the returned segments to print start time, end time, and text
- 3
Get the most out of it
- Use BatchedInferencePipeline with batch_size to speed up transcription on GPU, per the README benchmarks
- Use compute_type int8 or int8_float16 to reduce memory use
- Enable vad_filter=True to remove silence longer than 2 seconds by default
- Use word_timestamps=True when you need timing for each word
Downloads
Strengths
- The README reports up to 4x speed gains over openai/whisper at the same accuracy, with lower memory use
- Supports 8-bit quantization on both CPU and GPU to reduce resource needs
- Does not require FFmpeg to be installed separately, since audio decoding uses bundled PyAV libraries
Trade-offs
- GPU use requires separately installed NVIDIA cuBLAS and cuDNN libraries, which adds setup work
- Benchmarks in the README are specific to particular hardware and audio, so real results will vary
- The README is truncated and does not fully document all options; it points to source code for VAD parameters
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| faster-whisper (this) | 26.4k | Library | intermediate | 5 days ago | Python developers who need to add speech-to-text to an application |
| whisper | 112k | AI model / agent | intermediate | 2 months ago | Developers who want to add transcription of audio files to a Python project |
| whisper.cpp | 54.9k | Library | intermediate | 7 days ago | Developers who want to embed offline speech-to-text in a C, C++, or other native app |
| nano-vllm | 16.3k | Library | intermediate | 6 months ago | Developers who want to understand how LLM inference engines work internally |
| openwhispr | 10.5k | App | beginner | 4 days ago | People who want dictation in any app without sending audio to a cloud service |
| FunASR | 20.9k | Library | intermediate | 4 days ago | Developers who want to transcribe audio in Python with a pretrained model |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.