whisper.cpp

Use it when

• You need offline on-device speech transcription without installing a Python or PyTorch environment
• You want a fast local CLI or self-hosted HTTP server that runs Whisper models on CPU, Apple Silicon, or GPU
• You need word-level timestamps, confidence coloring, or karaoke-style subtitle output from audio files

What it solves

Not the fit when

• Training or fine-tuning models; the retained README states inference only
• The default whisper-cli path accepts 16-bit WAV input; convert other media first or build the optional FFmpeg decoding support
• The whisper-stream microphone example is described as naive and requires SDL2
• Word-level timestamps and confidence coloring are experimental features
• Speaker segmentation beyond the experimental tinydiarize support
• production speaker diarization
• text to speech generation
• model training or fine-tuning

Install

git clone https://github.com/ggml-org/whisper.cpp && cd whisper.cpp && sh ./models/download-ggml-model.sh base.en && cmake -B build && cmake --build build -j --config Release

Invoke

./build/bin/whisper-cli -m models/ggml-base.en.bin -f audio.wav (or run whisper-server for an HTTP API, whisper-stream for live mic input)

Alternatives

No reviewed alternatives recorded yet.