Voice, dictation & realtime
openai/whisper
Local speech recognition, the de facto standard
- 107k
- MIT
- Python
What it is and why it matters
Whisper is OpenAI's speech recognition model, widely used as a standard for transcription, including running locally. In one video it is integrated to give an agent voice understanding: they note the base model is around 600 MB and runs on CPU, even on somewhat older laptops. In another it is used as a transcription tool to measure the degradation of audio generated by video models. It is speech recognition (speech-to-text), not voice synthesis.
Key points
- The base model is small (around 600 MB) and runs on CPU
- Used to transcribe speech to text locally, no cloud
- Treated in the videos as a reference for transcription
Where we saw it
No place like localhost 2026-07-17 With the open code voice CLI, Whisper transcribes speech locally: the base model is around 600 MB and runs on CPU, even on somewhat older laptops. Local AI voice cloning Alex Ziskind 2026-06-18 They use Whisper as a transcriber to compute word error rate and thereby measure how much video-model audio degrades under quantization. This Is What Happens When You CRUSH An AI Video Model
Same category