speech.md

Voice transcription for macOS that runs entirely on your device. No server, no account, no audio leaving your Mac.

Built on SpeechAnalyzer and SpeechTranscriber (Apple Speech), with ScreenCaptureKit to capture what the other people in a call are saying.

What it does

Dictate — a global hotkey in any app: hold it, speak, release, and the text is pasted wherever your cursor is. A short tap keeps it listening until the next tap.

Meetings — records a meeting on two separate channels, your microphone and the system audio, each with its own analyzer. Both transcripts stream in live, side by side.

Files — drop in an audio file and get a transcript, along with how many times faster than real time it ran.

Dictionary — voice shortcuts: say "my email" and your full address comes out.

Markdown — optional. An on-device model (Foundation Models) reformats the dictation before pasting, turning spoken lists into bullets and fixing punctuation.

Performance

Measured on an Apple M5, macOS 26.5, pt-BR locale, low-latency mode. 27.7s of audio generated with say, three runs.

Measure	Result
File transcription	75× real time (27.7s of audio in 0.37s)
Markdown formatting, 14 words	0.62s (0.43–0.93)
Markdown formatting, 56 words	1.02s (0.90–1.12)

Formatting only runs when enabled, and it runs after transcription — it adds latency between releasing the hotkey and the text appearing, not while you speak.

The real-time figure measures file throughput, not the live experience. The targets for sustained use — first partial under 300 ms, sustained lag under 600 ms — have not been measured over a long real session yet.

Requirements

• Apple Silicon
• macOS 26 or later
• Xcode 26 or later

Build

./scripts/bundle.sh
open dist/speech.md.app

Signing

The script signs with the identity in SPEECH_SIGN_IDENTITY (default: speech.md Local). This matters more than it looks: macOS ties privacy permissions to the app's signature, and an ad-hoc signature (codesign --sign -) produces a new hash on every build — the system treats each rebuild as a different app and asks for microphone, screen and accessibility access again, every time.

To create a stable local identity, once:

openssl req -x509 -newkey rsa:2048 -keyout key.pem -out cert.pem -days 3650 \
  -nodes -subj "/CN=speech.md Local" \
  -addext "extendedKeyUsage=codeSigning" \
  -addext "basicConstraints=critical,CA:false"
openssl pkcs12 -export -legacy -out cert.p12 -inkey key.pem -in cert.pem \
  -passout pass:yourpassword -name "speech.md Local"
security import cert.p12 -k ~/Library/Keychains/login.keychain-db \
  -P yourpassword -T /usr/bin/codesign
security add-trusted-cert -r trustRoot -p codeSign \
  -k ~/Library/Keychains/login.keychain-db cert.pem

Or export SPEECH_SIGN_IDENTITY with an Apple Development identity you already have.

Permissions

Permission	Why
Microphone	Transcribe your voice
Screen Recording	Capture audio from the other participants
Accessibility	Paste dictated text into the focused app

Screen Recording is only needed for the "Others" channel. Turn off "Áudio dos outros participantes" in Settings and meetings run on the microphone alone, without that permission.

Architecture

Sources/SpeechMD/
├── App/          entry point
├── Speech/       SpeechPipeline, system audio capture
├── Dictation/    global dictation and text insertion
├── Meetings/     two-channel meeting recording
├── Files/        file transcription
├── Snippets/     voice shortcut dictionary
├── Formatting/   on-device Markdown pass
├── Island/       floating indicator by the notch
├── Settings/     preferences, global hotkey, permissions
└── Notetaker/    interface shell

Partial transcription never waits on formatting, summarization or persistence. Those consumers receive events and can drop work when they fall behind.

Status

Personal project, work in progress. Meetings and dictations live in memory — closing the app discards them. Persistence is next.

The interface is in Brazilian Portuguese.

License

MIT — see LICENSE.