Skip to main content
kombify SpeechKit is the speech-to-text (STT) and text-to-speech (TTS) framework that powers voice interaction in kombify AI. It runs as a standalone service with a Go backend and React frontend, connecting to your Companions and the Voice Agent pipeline.
SpeechKit lives in its own repository: kombify-SpeechKit. It is a separate component from the main kombify-AI service.

​
Two modes of operation

SpeechKit provides two distinct voice interaction modes:
Standard STT/TTS pipeline for kombify AI Companions.Flow: Microphone input → STT transcription → Companion processes request → TTS speech output
  • Push-to-talk or voice activity detection (VAD)
  • Audio visualization with waveforms
  • Works with any configured Companion
  • Provider hot-switching (change STT provider without restart)

​
Supported STT providers

SpeechKit supports six speech-to-text providers. You can switch between them at runtime without restarting the service.
ProviderTypeLatencyNotes
Azure SpeechCloudLowMicrosoft Cognitive Services
Google Cloud STTCloudLowGoogle Cloud Speech-to-Text
DeepgramCloudVery lowOptimized for real-time streaming
OpenAI Whisper APICloudMediumOpenAI hosted Whisper
Groq WhisperCloudLowGroq-accelerated Whisper inference
Faster WhisperLocalMediumRuns entirely on your machine, no cloud needed
For a fully local setup with no cloud dependencies, use Faster Whisper for STT and Qwen3-TTS for TTS.

​
Text-to-speech

SpeechKit uses Qwen3-TTS via the Kokoro pipeline for text-to-speech. This runs locally on your machine — no cloud API required.

​
Architecture

SpeechKit runs as a self-contained service with two components: 
┌──────────────────────────────────┐
│         React Frontend           │
│     (Vite 6 / React 19)         │
│  Audio capture, visualization,   │
│  provider selection UI           │
└──────────────┬───────────────────┘
               │ WebSocket / HTTP
┌──────────────▼───────────────────┐
│          Go Backend              │
│        (HTTP on :8787)           │
│  STT routing, TTS pipeline,      │
│  WebRTC (Voice Agent Mode)       │
└──────────────┬───────────────────┘
               │
    ┌──────────▼──────────┐
    │   STT Providers     │
    │   TTS (Qwen3/Kokoro)│
    │   Gemini Live API   │
    └─────────────────────┘
Tech stack:
  • Go 1.25 (backend HTTP server, WebSocket handling via gorilla/websocket)
  • React 19 + Vite 6 (frontend UI)
  • WebRTC for Gemini Live API connectivity

​
Configuration

​
API keys (BYOK)

SpeechKit uses the BYOK (Bring Your Own Keys) model. You provide API keys for whichever cloud STT providers you want to use.
1

Open the SpeechKit UI

Navigate to the SpeechKit frontend in your browser.
2

Select an STT provider

Choose your preferred provider from the dropdown. You can switch providers at any time without restarting.
3

Enter your API key

Provide the API key for the selected cloud provider. For Faster Whisper (local), no key is needed.

​
Provider selection

You can hot-switch between STT providers during a session. SpeechKit routes audio to the currently selected provider without requiring a restart.

​
Quick start

​
Assist Mode

1

Start SpeechKit

Launch the SpeechKit service. The Go backend starts on port 8787 and serves the React frontend.
2

Configure a provider

Select an STT provider and enter your API key (or choose Faster Whisper for local processing).
3

Start speaking

Use push-to-talk or enable VAD, then speak your request. The transcription is sent to your Companion, and the response is spoken back via Qwen3-TTS.

​
Voice Agent Mode

1

Configure Gemini API key

Voice Agent Mode requires a Google Gemini API key for the Gemini Live API.
2

Switch to Voice Agent Mode

Select Voice Agent Mode in the SpeechKit UI.
3

Start a conversation

Begin speaking. The Gemini Live API provides real-time bidirectional audio — you can interrupt and redirect the conversation naturally.

​
Platform support

PlatformStatus
WindowsProduction-ready
LinuxPlanned
SpeechKit is currently Windows-first. Linux support is on the roadmap.

​
Further reading

Voice interaction

User-facing voice features in kombify AI

BYOK setup

Configure your own API keys for AI providers