STP — Speech to Prompt

Talk to Antigravity. Completely hands-free.

STP is an Antigravity extension that lets you create prompts by voice and fire them straight into your Antigravity chat. No typing, no copy-paste, no need to press send, you say it out loud. You talk, it listens, you can ask Gemini to clean it up, you can enhance it or you can send it straight away — all without touching the keyboard or mouse.

After the voice pipeline was working, I used the STP extension to build the rest of the extension — so there are some rough edges, but it works, it's useful, and it's being actively improved.

How It Works in Practice

Basic flow — dictate and send

You speak: "Build a staff availability page. Each staff member should be able to set their weekly schedule using a calendar-style UI. Use the existing card styling from the dashboard. Send the prompt."

STP picks up send the prompt, injects your prompt directly into Antigravity chat, no need to press send, and restarts the mic. Done.

Enhance flow — let Gemini polish it first

You speak: "I need a modal that pops up when a staff member tries to set their availability but hasn't selected their service category yet. It should block them and prompt them to save their categories first. Enhance this prompt."

Gemini rewrites it into a precise, production-ready spec. You see the enhanced version inline and say:

"Approve." → fires into Antigravity. "Reject." → restores your original. "Try again." → Gemini takes another pass.

Clean flow — strip noise before sending

Say something awkward mid-dictation? Someone walked in? No problem:

You speak: "Add rate limiting to the API — hold on, hey, yeah I'll be right there — okay where was I — add a 429 response with a Retry-After header. Clean it up. Send the prompt."

Clean it up runs Gemini with a strict strip-only pass: fillers, interruptions, and off-topic noise are removed. Your actual intent is preserved. Then send the prompt fires it.

Pause and resume — mid-session

You speak: "Build a login page with Google OAuth and — pause."

Mic mutes immediately. Buffer is preserved. Later:

"Resume." → mic wakes back up exactly where you left off.

What's Under the Hood

• Audio pipeline — audio captured locally, denoised, transcribed. No browser dependency, no cloud STT lock-in.
• Optional real-time mode — drops latency from ~5s to ~300ms with a free key. Highly recommended for voice commands.
• Prompt enhancement — say "enhance this prompt" and get a polished, context-aware rewrite inline
• Voice cleanup — say "clean it up" to strip filler words and noise without touching your intent
• Approve / Reject / Try Again — review enhancements with buttons or purely by voice
• Hands-free send — say "send it" or "send the prompt" — injected into Antigravity instantly, no click
• Smart pause/resume — mic mutes on "pause", wakes on "resume" — buffer preserved throughout
• Workspace context — reads your open files, active conversation, and workspace name for smarter enhancements
• Multi-memory context — layer past Antigravity conversations as supplementary context for richer elaborations
• Secure key storage — Gemini API key stored in VS Code SecretStorage, never in code or config files
• Built-in noise suppression — 3-stage FFmpeg filter chain strips low-frequency rumble before audio reaches Gemini

Voice Command Reference

Commands work mid-sentence — you don't have to say them alone. Say them naturally as part of your dictation.

Note: Commands are ~10x faster in Deepgram mode (~300ms) vs Gemini chunked mode (~5s).

Getting Started

1. Install FFmpeg

Windows (recommended via winget):

winget install ffmpeg

Or download from ffmpeg.org and add to your system PATH. Restart VS Code after installing.

2. Get a Gemini API Key

Get a key at aistudio.google.com — no credit card required. The free tier (~15 req/min) is enough for normal use.

3. Add your key to STP

Open the STP panel from the Activity Bar, click KEY, and paste your Gemini API key. Stored in VS Code SecretStorage — never in a file, never leaves your machine.

4. Start talking

Click the microphone button and start dictating. In Gemini mode, audio processes in 3-second chunks — transcript updates live as you speak.

5. (Recommended) Enable Deepgram for real-time transcription

By default STP uses Gemini for transcription (offline, no extra keys, ~4-7s latency). Enable Deepgram for ~300ms latency and instant voice command response — it's free and takes 2 minutes.

Click LIVE in the STP panel header and follow the onboarding.

Deepgram — Real-Time Transcription (Optional)

Deepgram is an optional, opt-in third-party service that drops transcription latency to ~300ms via real-time streaming. A free API key is all you need.

Click the LIVE button in the panel and follow the onboarding. Your key is stored in VS Code SecretStorage and only sent to Deepgram's API during recording.

Deepgram Data Usage

Based on publicly available Deepgram documentation (early 2026). See deepgram.com/privacy for the authoritative source.

Get a free Deepgram API key — Privacy policy

Memory Context — Multi-Conversation Selection

STP automatically detects your current Antigravity conversation (the most recently modified chat) and feeds it into every enhancement as the primary context.

If you need to pull in knowledge from other past chats — a previous session on the same project, a different workspace's conversation — click the context card at the top of the panel.

Extras are read-only — they inform the elaboration but cannot be modified or replace the primary auto-detected chat.

Platform Support

Settings

Audio Pipeline

Gemini Mode (Default)

Mic → FFmpeg → [highpass=f=80] → [afftdn=nf=-25] → [silencedetect] → WAV chunks
                  Remove rumble      FFT denoiser       VAD trigger
                                                              ↓
                                                       Gemini 2.5 Flash
                                                         (transcribe)
                                                              ↓
                                                      Live transcript UI

1. highpass=f=80 — strips sub-80Hz rumble (HVAC, desk vibration, USB hiss)
2. afftdn=nf=-25 — FFmpeg's FFT denoiser; estimates noise floor in first ~0.4s and subtracts it every frame
3. silencedetect=noise=-40dB:d=2.5 — triggers VAD after 2.5s of true silence (on the cleaned signal)
4. Audio is segmented into 3-second WAV chunks and sent to Gemini for verbatim transcription
5. Each chunk's transcript is appended live to the panel as it comes back

Deepgram Mode (Optional)

Mic → FFmpeg → raw s16le PCM @ 16kHz → WebSocket → Deepgram nova-2
                                                          ↓
                                                 Interim results ~300ms
                                                          ↓
                                                  Final words committed
                                                          ↓
                                                  Live transcript UI

No WAV files written. No chunks. Audio streams directly in real-time for near-instant feedback.

Built With

Privacy

STP does not collect, store, or transmit any personal data.

No telemetry. No analytics. No STP backend.

License

MIT © 2026 — View on GitHub