Whisper

Whisper is OpenAI's open-weight speech recognition model released in September 2022 — trained on 680,000 hours of multilingual and multitask audio from the web. The model supports automatic speech recognition (ASR) in 99 languages, speech translation to English, language identification, and voice activity detection — all in a single model. Whisper variants range from tiny (39M parameters, fast on CPU) to large-v3 (1.5B parameters, highest quality). Released under MIT license with weights on Hugging Face. Whisper became foundational to modern speech-AI applications: meeting transcription tools (Otter, Fireflies, Granola, Microsoft Teams), podcast indexing, accessibility features, voice interfaces, and the speech components of multimodal AI systems. Quantized variants (faster-whisper, whisper.cpp with GGML format) enable real-time speech recognition on consumer hardware. Available through OpenAI's API and self-hosted on countless platforms. AI governance, AI compliance, and AI risk management programs deploy Whisper for on-prem speech processing supporting responsible AI through local speech recognition that keeps audio content inside enterprise AI perimeters.

Centralpoint Integrates Whisper for Speech Processing: Oxcyon's Centralpoint AI Governance Platform connects to self-hosted Whisper for speech recognition alongside OpenAI, Gemini, Llama, and other models. Centralpoint meters every call, keeps prompts and skills on-prem, and embeds speech-enabled chatbots into your portals via one JavaScript line.


Related Keywords:
Whisper,,