Sandwich Architecture
A voice agent architecture composed of three cascaded stages: speech-to-text (STT), large language model inference, and text-to-speech (TTS). The structure is clear and modular, but its latency equals the sum of the time consumed by each stage.
Timeline (last 90 days)
The core idea of the sandwich architecture is to place a large language reasoning model (LLM) in the middle, with an STT model added in front for speech recognition and a TTS model added at the back for speech output.
The greatest advantage of the sandwich architecture is the freedom to choose the currently strongest reasoning model as the agent's brain, without being limited by whether the model itself supports voice.
This project achieved a response latency of approximately 0.3 seconds through fine-grained optimization of the front-end STT and back-end TTS.
The biggest pain point of the sandwich architecture is degraded real-time performance, and the core of successful deployment lies in whether the latency problem can be solved.
For the vast majority of enterprise scenarios, the sandwich architecture (STT + strong reasoning LLM + TTS) is currently the most pragmatic and controllable choice.
All Facts (5)
The core idea of the sandwich architecture is to place a large language reasoning model (LLM) in the middle, with an STT model added in front for speech recognition and a TTS model added at the back for speech output.
50%UnverifiedThe greatest advantage of the sandwich architecture is the freedom to choose the currently strongest reasoning model as the agent's brain, without being limited by whether the model itself supports voice.
50%UnverifiedThis project achieved a response latency of approximately 0.3 seconds through fine-grained optimization of the front-end STT and back-end TTS.
50%UnverifiedThe biggest pain point of the sandwich architecture is degraded real-time performance, and the core of successful deployment lies in whether the latency problem can be solved.
50%UnverifiedFor the vast majority of enterprise scenarios, the sandwich architecture (STT + strong reasoning LLM + TTS) is currently the most pragmatic and controllable choice.
50%