Skip to content

KalamAI

KalamAI is a real-time multilingual video conferencing platform built specifically for Africa.

Code Demo Cerebral Valley

Core trick

Chains self-hosted speech-to-text, Claude translation, and a three-tier TTS fallback so each speaker hears the meeting in their language.

Tags

educationvoiceaccessibility

Description

KalamAI is a real-time multilingual video conferencing platform built specifically for Africa. Every participant speaks in their native language (Hausa, Fulfulde, Arabic, French, Bambara) and hears the meeting translated live into their chosen language, voice-to-voice, with no configuration required. The full speech-to-speech pipeline runs on self-hosted infrastructure: faster-Whisper for transcription, Meta SeamlessM4T v2 for translation, and a three-tier TTS system (edge-tts for neural voices, Meta MMS-TTS for low-resource African languages, espeak-ng as universal fallback). Beyond real-time translation, KalamAI is architected as a sovereign communication layer for institutions. No audio or transcript data touches a third-party cloud. The platform runs entirely via Docker Compose on any server — including on-premise government hardware in Chad — and supports 10 languages at launch, with an active roadmap to integrate 16 Sara dialects spoken by ~3 million people in Chad and the CAR, through fine-tuned Whisper and VITS/MMS-TTS models trained on locally collected corpora.
abdelazizharane/KalamAI★ 00 forksTypeScriptCreated Apr 21, 2026 · last push Apr 26, 2026

README excerpt

# KalamAI - Visioconference platform for low-ressource languages 
A video conferencing platform with real-time audio translation powered by Claude Opus 4.7, designed for African and international institutions.

<img width="334" height="85" alt="image" src="https://github.com/user-attachments/assets/8ebfa359-ae39-4360-b172-fd8e3ed0df18" />



## 🚀 Quick Start

```bash
# 1. Clone the repo
git clone https://github.com/votre-username/kalamAI
cd kalamAI

# 2. Environment variables
cp .env.example .env
echo “ANTHROPIC_API_KEY=sk-ant-...” >> .env

# 3. Start everything
docker compose up --build
```

Frontend: http://localhost:3000  
API: http://localhost:8000  
API Docs: http://localhost:8000/docs  

## 🧠 Architecture

```
French speech  →  faster-whisper (STT)
           →  Claude Opus 4.7 (translation)
           →  CosyVoice3-0.5B (TTS 150ms) — FR/EN/ES/DE/IT/RU
           →  Fish Speech V1.5 (TTS ~200ms) — AR/HA/SW + 77 others
```

### 4 Claude Opus 4.7 Agents

| Agent | Role | Frequency |
|---|---|---|
| **Translation** | Translates each segment in real time | Continuous |
| **Summary** | Structured summary of the meeting | Every 5 min |
| **Action Items** | Extracts decisions and commitments | Every 2 min |
| **Email** | Drafts the post-meeting follow-up | End of meeting |

### Supported Languages

| Language | ASR (speech-to-text) | TTS (text-to-speech) |
|---|---|---|
| French | ✅ Whisper | ✅ CosyVoice3 (150ms) |
| English | ✅ Whisper | ✅ CosyVoice3 (150ms) |
| Standard Arab

Team