Faster-Whisper Engine

Faster-Whisper is a reimplementation of OpenAI’s Whisper model using CTranslate2, a fast inference engine for Transformer models. It is up to 4x faster than openai/whisper with significantly reduced memory footprint.

---

Key Advantages

• Universal Multilingual Support: 99+ languages with automatic language detection.
• Word-Level Timestamps: Precise alignment for subtitles (SRT/VTT).
• Quantization Support: Native int8, int8_float16, and float16 execution.

---

Configuration Example

ASR-Providers:
  faster-whisper:
    enable: true
    engine: faster-whisper
    load:
      model_name: medium # tiny, base, small, medium, large-v3
      model_path: models/faster-whisper-medium
      device: cuda # or cpu
      compute_type: float16 # or int8

---

Performance Benchmark

Model Size	Parameters	VRAM (fp16)	RTF (NVIDIA RTX 4090)
base	74M	1.0 GB	0.03x
small	244M	2.0 GB	0.06x
medium	769M	4.5 GB	0.12x
large-v3	1550M	8.0 GB	0.22x