Aurum¶

Speech both ways. On-device by default.

Private speech I/O on your machine:

• STT — audio → text (whisper.cpp local default; optional OpenRouter / OpenAI / xAI)
• Cleanup — post-transcript styles (on-device rules or OpenRouter LLM)
• TTS — text → mono WAV (local ONNX KittenTTS / Kokoro; optional OpenRouter / OpenAI / ElevenLabs / xAI)

The same core powers the aurum CLI, Rust embeds (aurum-core), and native embeds (aurum-ffi).

30-second path¶

git clone https://github.com/joe-broadhead/aurum.git
cd aurum
./scripts/install.sh
# or: cargo install aurum-stt

aurum models
aurum meeting.m4a --model tiny-q5_1
aurum meeting.m4a --cleanup clean
aurum tts "Hello from aurum" --output-file /tmp/hello.wav

Architecture¶

flowchart LR
  subgraph stt [STT]
    A[Audio / PCM] --> B[Load]
    B --> C{Provider}
    C -->|local| D[whisper.cpp]
    C -->|openrouter / openai / xai| E[Remote STT]
    D --> F[postprocess]
    E --> F
    F --> G[cleanup]
    G --> H[txt / srt / json]
  end
  subgraph tts [TTS]
    T[Text] --> U[validate]
    U --> V{Provider}
    V -->|local| W[Kitten / Kokoro ONNX]
    V -->|openrouter / openai / elevenlabs / xai| X[Remote TTS]
    W --> Y[mono WAV]
    X --> Y
  end

Workspace¶

Where to go next¶

Status¶

v0.0.23 (published tip) — correctness & integration patch after the independent v0.0.22 audit (SRT provenance, batch integrity, SDK/OpContext wiring, native SDK consumer matrix). See v0.0.23 acceptance.

Prior product-outcomes cut: v0.0.22 (JOE-2215 / JOE-2226). Continuous 0.0.x (not 1.0).