Build a Python voice assistant around DeepSeek R1
AssemblyAI’s lesson connects transcription, a local language model and speech synthesis.
A dated developer lesson
AssemblyAI published this Python walkthrough on February 11, 2025. It connects AssemblyAI speech recognition, DeepSeek R1 through Ollama and ElevenLabs speech synthesis. It suits readers interested in wiring these components into an application, with a spoken question becoming text before the model produces a response.
Prepare the separate services
The linked written companion is dated September 30, 2026. It requires AssemblyAI and ElevenLabs API keys, Python dependencies and a downloaded Ollama model. The language model runs locally, while transcription and speech synthesis use cloud APIs. This is therefore not a fully offline assistant.
Try it: a CurioFuse exercise
An original editorial exercise with a checkable outcome.
CurioFuse exercise: trace a harmless sentence
Use the invented spoken request “What is project Amber?” Draw four boxes: recorded speech, transcribed text, model response and synthesized speech. Under each, write where processing occurs according to the companion and what data crosses into the next box. Leave an unknown location marked unknown.
Your diagram is useful when every cloud call is visible and local model execution is not mistaken for an entirely offline workflow. Add one failure question per boundary, such as what happens when transcription returns the wrong project name.
