Field study · Local speech interface
Bilingual Voice Assistant
Whisper transcription, local language models, and offline speech combined into an English and Spanish interface with editable memory.
Speech is one part of the system.
- 01Capture an audio turn
- 02Transcribe locally with Whisper
- 03Route the session and language
- 04Retrieve relevant memory
- 05Generate a local model reply
- 06Speak with an offline voice
The implementation separates transcription, model replies, speech synthesis, person/session context, and feedback. Memory can be reviewed, edited, or forgotten.
Make new capabilities explicit.
Typed input is implemented in the development lane. The inspected roadmap records it as awaiting promotion after user evaluation. Automatic speaker identification and room-scale listening distance are not presented as established capabilities.
This project is distinct from Dictation Coach, whose job is to transcribe, optionally rewrite, and type speech into an active application.
A local project, not a public conversation endpoint.
This case study describes the architecture. It does not expose the household server, recordings, conversation logs, or personal memories.