1️⃣ whisper.cpp (54.8k stars)

OpenAI's Whisper in plain C/C++ with no dependencies. Runs on Mac, Windows, Linux, iOS, Android, WebAssembly and even a Raspberry Pi. MIT.

https://github.com/ggml-org/whisper.cpp

2️⃣ faster-whisper (26.3k stars)

Whisper rebuilt on CTranslate2. Its team says up to 4x faster than openai/whisper for the same accuracy, with less memory. MIT.

https://github.com/SYSTRAN/faster-whisper

3️⃣ WhisperX (24.9k stars)

Adds word-level timestamps and speaker diarization, so you know who said what and when. Built on the faster-whisper backend.

https://github.com/m-bain/whisperX

4️⃣ sherpa-onnx (15.6k stars)

A complete offline speech toolbox: recognition, synthesis, diarization, voice activity detection and keyword spotting, on mobile, desktop and the web. Apache 2.0.

https://github.com/k2-fsa/sherpa-onnx

5️⃣ Vosk (15.2k stars)

Offline recognition for 20+ languages with small models of about 50 MB and a streaming API for live results. Apache 2.0.

https://github.com/alphacep/vosk-api

6️⃣ SenseVoice (9.6k stars)

A speech model that also detects the language, the emotion in the voice and audio events. MIT.

https://github.com/FunAudioLLM/SenseVoice

Transcription inside your own app? 1️⃣. Faster Whisper in Python? 2️⃣. Meetings with several speakers? 3️⃣. A full voice stack on a phone? 4️⃣. Live results on small hardware? 5️⃣. Emotion on top of the words? 6️⃣.