Private Speech-to-Text — Your Voice Never Leaves Your Device
No cloud processing. No data collection. No accounts. lowercase runs entirely on your hardware, and your voice recordings are never stored or transmitted.
The privacy problem with dictation
When you use Google Voice Typing, Apple's Siri-based dictation, or Microsoft's speech services, your audio is sent to company servers for processing. These companies may retain your voice recordings for quality improvement, use them to train machine learning models, or store transcripts alongside your account data.
Even services that claim "on-device processing" often send data to the cloud for enhanced accuracy or fall back to server processing in certain conditions. The privacy policies are long, the opt-out settings are buried, and the actual data flow is opaque.
For anyone dictating sensitive information — medical notes, legal documents, personal journal entries, confidential business communications — this is an unacceptable risk. You shouldn't have to trust a corporation's privacy policy to use your own voice.
How lowercase guarantees privacy
lowercase's privacy isn't a policy promise — it's an architectural guarantee. The app contains no networking code for speech processing. There is no server to send data to, no API endpoint to call, and no analytics SDK collecting usage data.
No servers
lowercase has no backend infrastructure. There are no servers to receive your audio, store your transcripts, or log your usage. The app is entirely self-contained on your device.
No telemetry
No analytics, no crash reporting, no usage tracking. lowercase doesn't know how often you use it, what you dictate, or even that you've installed it.
No accounts
No sign-up, no email, no login. There's no user identity associated with your usage. You download, install, and use — that's it.
No audio storage
Audio exists only in temporary memory during active transcription. Once text is produced, the audio buffer is discarded. No recordings are saved to disk.
What cloud dictation services do with your voice
Understanding what happens to your audio with mainstream services helps illustrate why on-device processing matters.
Private dictation for sensitive work
Some professions handle information that must never leave a controlled environment. Cloud-based dictation is a non-starter for these use cases. lowercase's fully on-device architecture makes it suitable where other dictation tools aren't.
Medical professionals
Dictate patient notes, observations, and clinical documentation without PHI ever touching a third-party server. Supports HIPAA-conscious workflows where data must stay on-premises.
Legal professionals
Draft briefs, contracts, and privileged communications by voice. Attorney-client privilege and work product doctrine demand that sensitive content stays within your control.
Journalists & researchers
Transcribe interviews with confidential sources, draft investigative reporting, and process sensitive research data — all without exposing your work to cloud services.
Personal journaling
Your private thoughts deserve private tools. Dictate journal entries, therapy notes, and personal reflections knowing that no one — not even the app developer — can access them.
Business & executive
Draft confidential memos, board communications, M&A notes, and strategic plans. Corporate data stays on corporate hardware.
Government & defense
Environments where data classification prohibits cloud transmission. On-device processing meets air-gapped and restricted-network requirements.
The technical architecture of privacy
Privacy claims are only as strong as their technical implementation. Here's exactly what happens when you dictate with lowercase:
Audio capture
Your microphone captures audio into a local memory buffer. This buffer is held in RAM only — nothing is written to disk.
On-device inference
The NVIDIA Parakeet TDT model, stored locally on your device (~500 MB), processes the audio buffer. On Mac and iOS, inference runs on Apple Neural Engine. On Android, the system SpeechRecognizer handles processing.
Text output
The model produces transcribed text, which is placed on your system clipboard and pasted into your active application.
Cleanup
The audio buffer is released from memory. No audio file, temporary cache, or log is created at any point in the pipeline.
This pipeline has no step where data leaves your device. There is no "opt-in to cloud processing," no "send diagnostics" toggle, no "improve dictation" setting that silently uploads recordings. The privacy is structural.
More about lowercase
Your voice. Your device. Nobody else.
Download lowercase for truly private dictation — free, open source, and architecturally incapable of collecting your data.