Reverbal Server

Reverbal generates insights for your verbal conversation as it happens, helping you close that important sales call or discussing travel plans with your friend.

This server utilises Whisper and ChatGPT APIs to transcribe your audio and generate insights respsctively. You can adjust how ChatGPT generates insights by sending a prompt prior to starting the audio stream from your client.

Run

Pull docker image:

docker pull birudeghi/continuousgpt:beta

Use your own OpenAI token in docker-compose.yaml:

OPENAI_KEY=<your-own-key-here>

Run docker:

docker-compose up -d

API Reference

This server works on a simple WebSocket connection that accepts different types of events.

To Server

WebSockets ws://localhost:80/

prompt event

Sets a prompt to be sent together with your transcribed text.

{ 
 "event": "prompt",
 "prompt": "This is a message to customers in a call centre: "
}

media event

Audio streams in chunks that will be aggregated before being transcribed upon stop or break event.

{ 
 "event": "media",
 "sequenceNumber": "3", 
 "media": "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" 
}

break event

Aggregates collected audio streams to be transcribed. Does not close the connection.

{
  "event": "break"
}

stop event

Aggregates collected audio streams to be transcribed and then closes the connection.

{
  "event": "stop"
}

From Server

WebSockets ws://localhost:80/

Stream response

{
  "text": "",
  "stream": "start"
}
------
{
  "text": "This is the nature of...",
  "stream": "streaming"
}
------
{
  "text": "",
  "stream": "stop"
}

Error response

{
  "error": "There's a problem..."
}

Debug

To debug, create an .env file in root for your OpenAI token and run:

python app.py