Abdullah
12/17/2025, 1:34 PMVapi Support Bot
12/17/2025, 1:36 PM'ar', but automatic switching between Arabic and English is not yet available. For best results, configure your assistant to use Deepgram's 'whisper' model with Arabic as the language:
json
"transcriber": {
"provider": "Deepgram",
"model": "whisper",
"language": "ar"
}
If you need true bilingual auto-detection, you may need to monitor VAPI updates or consider a custom transcriber solution.
Reference: [VAPI Changelog April 25, 2025](https://docs.vapi.ai/changelog/2025/4/25)
Source:
- [Changelog April 25, 2025](https://docs.vapi.ai/changelog/2025/4/25)Abdullah
12/17/2025, 1:45 PMChiranjeet Mishra
12/18/2025, 1:54 PMws, express, dotenv, and your transcriber’s SDK (e.g., Deepgram).
2. Configure the Environment:
- Create a .env file with your transcriber’s API key and server port.
3. Create Transcription Service:
- Implement a service in transcriptionService.js to handle the audio data and interact with your transcriber.
4. Set Up WebSocket Server:
- Use express and ws to create a server that listens for connections from Vapi, processes audio, and returns transcription results.
5. Configure and Test:
- Deploy and test your custom server. Expose it using a tool like ngrok for secure connection.
For handling bilingual transcription (Arabic and English), Vapi currently lacks automatic switching capabilities. If this is crucial, you could develop a custom endpoint with logic to switch transcribers based on detected language segments. Monitoring updates on Vapi's capabilities or employing a solution like Vapi Squads, which allows using different assistants for different languages, might be beneficial. For comprehensive guidance on switching techniques, refer to [Vapi's documentation on multilingual support](https://docs.vapi.ai/customization/multilingual).
Feel free to reach out if you have further questions on setting this up.