TiltedVoice
Voice-to-text platform with two packages — one for global usage via on-device Whisper, and one for Indian languages via Sarvam AI cloud. Two packages, one mission: voice everywhere.
Two Packages
International
@tiltedprompts/voiceOn-device transcription via OpenAI Whisper. Zero cloud dependency. 99 languages. Sub-500ms latency. Fully offline.
Indic
@tiltedprompts/voice-indicCloud transcription via Sarvam AI. 22 Indian languages. Code-switching support. Voice agent builder.
@tiltedprompts/voice
On-device transcription powered by OpenAI Whisper. Runs entirely on your machine with Metal GPU acceleration on Apple Silicon. No audio data leaves your device.
On-Device Processing
Whisper runs locally. No cloud dependency, no latency penalty, no privacy concerns.
Push-to-Talk & Toggle
Two recording modes — hold-to-speak for quick commands, or toggle on/off for longer dictation.
99 Languages
Full Whisper language coverage including CJK, Arabic, Hebrew, and all European languages.
Custom Dictionary
Add domain-specific words, project names, and technical terms for higher accuracy.
Whisper Models
| Model | Size | Speed | Best For |
|---|---|---|---|
| tiny | 75 MB | ~32x realtime | Drafts, quick notes |
| base | 142 MB | ~16x realtime | Casual dictation |
| smallrecommended | 466 MB | ~6x realtime | Good accuracy |
| medium | 1.5 GB | ~2x realtime | High accuracy |
| large-v3 | 3.1 GB | ~1x realtime | Maximum accuracy |
Installation
# Desktop app (macOS, Windows, Linux)
brew install --cask tilted-voice # macOS
# npm package for developers
npm install @tiltedprompts/voice
# Quick CLI usage
npx tilted-voice --model small --lang en
@tiltedprompts/voice-indic
India-specific voice AI powered by Sarvam AI. Build multilingual voice agents that work over phone calls, WhatsApp, and web. The only managed platform for Indian-language voice automation.
22 Indian Languages
Hindi, Tamil, Telugu, Bengali, Kannada, Marathi, Gujarati, Malayalam, Punjabi, Odia, and 12 more.
Code-Switching
Handles Hinglish, Tanglish, and other mixed-language conversations natively. No pre-selection needed.
Telephony Integration
Exotel for Indian phone numbers (toll-free, local DID). Inbound call routing to voice agents.
WhatsApp Business Calling
Voice agents on WhatsApp via Gupshup BSP. Same agent handles text and voice calls.
Supported Indian Languages
Voice Agent Pipeline
The Indic package provides a complete voice agent pipeline — speech-to-text, language model processing, and text-to-speech — all optimized for sub-800ms end-to-end latency.
Sarvam Saaras v3
Speech → Text
GPT-4.1-mini
Reasoning
Sarvam Bulbul v3
Text → Speech
End-to-end latency: <800ms via WebRTC (LiveKit)
Installation
# npm package
npm install @tiltedprompts/voice-indic
# Python package
pip install tiltedvoice
# REST API
curl -X POST https://api.tiltedprompts.com/voice/v1/transcribe \
-H "Authorization: Bearer $TILTED_API_KEY" \
-F "audio=@recording.wav" \
-F "language=hi"
Tech Stack Comparison
| Layer | International | Indic |
|---|---|---|
| STT | Whisper (on-device) | Sarvam Saaras v3 (cloud) |
| TTS | — (dictation only) | Sarvam Bulbul v3 (cloud) |
| LLM | — (no reasoning) | GPT-4.1-mini (agent brain) |
| Transport | Local audio capture | WebRTC via LiveKit |
| Runtime | Rust/Tauri (desktop) | Python FastAPI (backend) |
| Telephony | — | Exotel (India) + Twilio (global) |
| — | Gupshup BSP |
Next: TiltedCode
Agent-optimized production templates