Convert English speech to text securely. Everything runs offline in your browser using a highly compressed AI, guaranteeing total privacy for sensitive meetings and memos.
By using WebAssembly, the transcription engine runs inside your computer's RAM. Confidential interviews, medical dictations, and private memos never leave your device.
Because there is no remote server processing your file, you aren't subjected to arbitrary tier limits, bandwidth throttling, or expensive transcription subscriptions.
To ensure the AI is lightweight enough to run securely inside your browser without crashing your computer, we utilize a specialized, highly-compressed English-only model.
Absolutely not. We consider audio uploads to be a massive security risk. Tangent downloads the transcription "brain" (the AI model) to your computer and processes the audio locally in your browser.
The AI language model is approximately 70MB. On your very first use, your browser must securely download this file. It is then saved into your local cache, meaning all future transcriptions will begin instantaneously—even if you disconnect from the internet.
Since you are using your own computer's CPU to process the audio, there are no artificial time limits or paywalls. However, very long files will take a proportionally longer amount of time to process.
Currently, this tool supports English only. By restricting the language, the AI model is kept small enough (~70MB) to download quickly and run locally on your hardware. Multilingual models are often too massive for stable browser execution.
Because this runs 100% offline in your browser, it uses a highly compressed "Tiny" AI model. It is incredibly fast and highly accurate for clear voice memos, podcasts, and structured meetings. However, it may occasionally struggle with thick accents, extreme background noise, or heavy overlapping speakers compared to massive, cloud-based commercial APIs.