Whissle Gateway lets businesses run voice AI—including speech recognition, speaker identification, and sales coaching analysis—entirely on their own servers in a 500MB Docker container.

What happened

Whissle released a containerized voice AI system that performs automatic speech recognition (ASR), text-to-speech (TTS), speaker diarization, and AI-powered call analysis locally without cloud dependency. The system downloads ~2 GB of models on first run and caches them thereafter, and supports multiple language variants including English, Hindi-English, Mandarin, and 23 languages total.

Why it matters

Businesses handling sensitive customer calls—such as sales teams, debt collections, and interviews—can now analyze conversations for coaching and compliance without sending audio to third-party APIs. The system extracts metadata per segment (emotion, behavior, speaker age/gender, intent) in a single forward pass, and can summarize transcripts using Claude or Gemini with custom prompts, letting organizations keep call data on premises.

What to watch

The system runs on both CPU (laptop/MacBook at 1–3 concurrent calls) and GPU (A100 40GB at 50–80 concurrent calls), with variants optimized for speed (en-lite, ~500 MB) or quality (en-full, ~2 GB). It is available immediately via Docker image (whissleasr/whissle-gateway:latest) with local authentication and REST/WebSocket APIs.