faster-whisper 命令

常用示例

说明

faster-whisper is a reimplementation of OpenAI's Whisper using CTranslate2, a fast inference engine for Transformer models. It provides up to 4x faster transcription than the original Whisper while using less memory. The tool supports all Whisper model sizes. Larger models are more accurate but slower. The compute type parameter controls precision: int8 is fastest and most memory-efficient, float16 is a good balance on GPU, float32 is highest precision. Voice activity detection (VAD) filtering skips silent sections, improving both speed and accuracy. Language detection is automatic but specifying the language avoids detection overhead. The base library is installed via `pip install faster-whisper` (Python API only). For CLI usage, install a wrapper such as `pip install faster-whisper-cli` or `pip install whisper-ctranslate2`. CTranslate2 handles model conversion automatically. GPU acceleration requires CUDA toolkit.

参数

--model _SIZE_

Model size: tiny, base, small, medium, large-v1, large-v2, large-v3 (default: small).

--language _LANG_

Language code (en, de, fr, etc.) or auto-detect.

--task _TASK_

Task: transcribe or translate.

--output_format _FORMAT_

Output format: txt, vtt, srt, tsv, json, all.

--output_dir _DIR_

Output directory for results.

--word_timestamps _BOOL_

Include word-level timestamps.

--device _DEVICE_

Device: cpu, cuda, auto (default: auto).

--compute_type _TYPE_

Compute type: int8, float16, float32 (default: int8 on CPU).

--beam_size _N_

Beam search size (default: 5).

--vad_filter _BOOL_

Enable voice activity detection filter (uses Silero VAD).

--initial_prompt _TEXT_

Optional text to provide as initial prompt for the decoder.

--threads _N_

Number of CPU threads.