Skip to main content

Local AI voice processing platform: audio/video → knowledge base, global voice input

Project description

Aureka

aural + eureka — 聽到,即發現知識

本機 AI 語音處理平台,兩個核心使用模式:

模式 說明 觸發方式
批次處理 影片/音訊 → 結構化 Markdown,可丟入知識庫 aureka process
語音輸入 全域熱鍵 → 說話 → 文字出現在游標位置 常駐 daemon + 熱鍵

安裝

PyPI(推薦)

# 基本安裝(daemon + LLM client)
pip install aureka

# 按需加裝功能模組
pip install "aureka[asr]"           # ASR(faster-whisper)
pip install "aureka[tts]"           # TTS(Kokoro)— 僅 Linux / macOS
pip install "aureka[batch]"         # 批次流水線(需另裝 ffmpeg,見下方)
pip install "aureka[voice]"         # 語音輸入 client(pynput/pystray)
pip install "aureka[all]"           # 以上全部(Windows 請用下方指令)

注意:PyTorch 需依平台單獨安裝(見下方),不包含在 extras 中。

Windows 用戶:Kokoro TTS 目前無 Windows wheel,請跳過 [tts]

pip install "aureka[asr,batch,voice]"

從原始碼安裝

git clone https://github.com/fcwu/aureka
cd aureka
pip install -e ".[all]"
pip install -r requirements-dev.txt   # 測試用

Python 版本

需要 Python 3.11 或 3.13(推薦)。Python 3.14+ 目前許多 ML 套件尚未支援,請勿使用。

PyTorch(依平台)

平台 指令
NVIDIA GPU(Linux / Windows) pip install torch --index-url https://download.pytorch.org/whl/cu121
AMD GPU(僅 Linux,ROCm 不支援 Windows) pip install torch --index-url https://download.pytorch.org/whl/rocm6.1
Apple Silicon / CPU only pip install torch

Windows 用戶:只支援 NVIDIA CUDA 或 CPU。若不確定,直接 pip install torch 即可(CPU 模式)。

ffmpeg(批次處理必要)

# Ubuntu / Debian
sudo apt install ffmpeg

# Fedora / RHEL
sudo dnf install ffmpeg

# macOS
brew install ffmpeg

# Windows
winget install ffmpeg
# 或:choco install ffmpeg

設定檔

cp config.example.toml config.toml
# 編輯 config.toml:填入 LM Studio / Ollama 端點

最少需要設定 [llm][vlm]base_url,其他欄位有預設值。


批次處理

用法

# 處理影片(提取音訊 + 關鍵畫面 + ASR + VLM + LLM 摘要)
aureka process lecture.mp4

# 處理音訊(只有 ASR + LLM 摘要,無畫面分析)
aureka process podcast.mp3

# 自訂參數
aureka process video.mp4 --frame-interval 60 --device cuda --output-dir ~/notes/inbox

輸出

結果寫入 output/YYYYMMDD-<slug>.md,格式如下:

---
source: video
original_file: lecture.mp4
duration: 45:32
processed_at: 2026-05-01T14:30:00
---

# <自動萃取的標題>

## 摘要
## 重點
## 逐段紀錄
## 視覺內容
## 原始轉錄

完成後可直接丟入 mykb inbox/ 走 triage → ingest 流程。


TTS 回讀

# 直接朗讀文字
aureka speak "今天的工作重點是什麼"

# 朗讀 Markdown 檔案(自動略過 frontmatter 和標記語法)
aureka speak --file path/to/note.md

# 存成 WAV 不播放
aureka speak "測試" --output out.wav

語音輸入(Typeless-like)

啟動 Daemon

# 啟動常駐 daemon(預載 ASR 模型,避免每次冷啟動)
aureka daemon start

# 確認狀態
aureka daemon status
# → Daemon: running (PID 12345) → http://127.0.0.1:7777

# 停止
aureka daemon stop

Daemon log:/tmp/aureka-daemon.log

啟動語音輸入 Client

aureka type            # 預設 refine 模式
aureka type --mode transcribe   # 直接轉錄,不過 LLM
aureka type --mode translate --lang en   # 說中文,輸出英文

或啟動系統托盤 client(有 GUI 圖示,可右鍵切換模式):

python -m aureka._daemon_serve --host 127.0.0.1 --port 7777 &
python -c "from aureka.client import start_tray; start_tray()"

錄音模式(config.toml)

[hotkey]
trigger    = "<ctrl>+<alt>+space"
mode       = "hold-to-record"   # hold-to-record / toggle / vad
input_mode = "refine"           # transcribe / refine / translate
lang       = "zh"
模式 說明
hold-to-record 按住熱鍵錄音,放開停止(預設)
toggle 按一下開始,再按停止
vad 偵測靜音自動停止

AI 後處理模式

模式 說明 額外延遲
transcribe 直接注入轉錄文字 0
refine 去除語氣詞、修正語法 +1–2s
translate 翻譯成指定語言 +1–2s

快速測試(不需真實 GPU 或模型)

Step 1:生成測試音訊

python tests/scripts/gen-test-audio.py
# → tests/fixtures/silence-1s.wav
# → tests/fixtures/speech-zh.wav

Step 2:啟動 mock LLM server

python tests/scripts/mock-llm-server.py --port 11434 &
# 模擬 /v1/chat/completions(含 vision)和 /v1/models

Step 3:啟動 daemon(測試模式,跳過模型載入)

AUREKA_TEST_MODE=1 AUREKA_CONFIG=tests/config.test.toml aureka daemon start
curl http://127.0.0.1:7777/health
# → {"status":"ok","version":"0.1.0"}

Step 4:測試 WebSocket 語音輸入

python tests/scripts/ws-client-test.py \
  --audio tests/fixtures/speech-zh.wav \
  --mode transcribe

# 預期輸出:
# [←] {"type": "transcript", "text": "[mock transcript]", "final": true}
# [←] {"type": "done"}
python tests/scripts/ws-client-test.py \
  --audio tests/fixtures/speech-zh.wav \
  --mode refine

# 預期輸出:
# [←] {"type": "transcript", ...}
# [←] {"type": "refined", "text": "這是一段經過整理的文字。", "final": true}
# [←] {"type": "done"}

Step 5:測試批次處理

AUREKA_TEST_MODE=1 AUREKA_CONFIG=tests/config.test.toml \
  aureka process tests/fixtures/silence-1s.wav --output-dir /tmp/aureka-out
# → /tmp/aureka-out/YYYYMMDD-silence-1s.md

執行測試

# 全部測試(unit + integration + e2e)
pytest tests/ -v

# 只跑 unit(快,無外部相依)
pytest tests/ -v -m unit

# 只跑 integration(需 mock LLM server,由 conftest 自動啟動)
pytest tests/ -v -m integration

# 只跑 e2e(啟動真實 daemon 子程序)
pytest tests/ -v -m e2e

專案結構

aureka/
├── aureka/
│   ├── __main__.py       # CLI 入口(process / speak / type / daemon)
│   ├── config.py         # config.toml 載入(AUREKA_CONFIG env var)
│   ├── device.py         # 裝置偵測(cuda / mps / cpu)+ ASR 後端選擇
│   ├── asr.py            # ASR 統一介面(TheWhisper / faster-whisper)
│   ├── llm.py            # LLM / VLM 呼叫(OpenAI-compatible)
│   ├── tts.py            # Kokoro TTS 封裝 + Markdown 前處理
│   ├── pipeline.py       # 批次流程編排
│   ├── daemon.py         # FastAPI daemon(WebSocket /ws + HTTP)
│   ├── recorder.py       # 麥克風錄音(hold / toggle / VAD)
│   ├── hotkey.py         # 全域熱鍵(pynput)
│   ├── client.py         # 語音輸入 client(pystray + WebSocket)
│   ├── injector.py       # 文字注入(xdotool / 剪貼簿)
│   ├── ffmpeg_utils.py   # 音訊提取 + 關鍵畫面截取
│   └── formatter.py      # Markdown 輸出格式化
├── tests/
│   ├── conftest.py               # 共用 fixtures(mock server、config)
│   ├── test_device.py            # unit: 裝置偵測、ASR 後端選擇
│   ├── test_tts.py               # unit: Markdown 前處理
│   ├── test_injector.py          # unit: 文字注入邏輯
│   ├── test_llm.py               # integration: LLM/VLM client
│   ├── test_pipeline.py          # integration: 批次流水線
│   ├── test_daemon.py            # integration: HTTP + WebSocket
│   ├── test_e2e.py               # e2e: daemon 程序管理
│   ├── fixtures/                 # 測試音訊(gen-test-audio.py 生成)
│   └── scripts/
│       ├── gen-test-audio.py     # 生成測試 WAV fixtures
│       ├── mock-llm-server.py    # mock OpenAI-compatible server
│       └── ws-client-test.py     # WebSocket 手動測試工具
├── docs/
│   └── design.md
├── config.example.toml   # 設定範本
├── requirements.txt
└── requirements-dev.txt

平台支援

平台 語音輸入 批次處理 ASR 加速 TTS 加速
NVIDIA Linux CUDA (TheWhisper) CUDA (Kokoro)
AMD Linux ROCm (faster-whisper) ROCm (Kokoro)
Apple Silicon CoreML (TheWhisper) MPS (Kokoro)
CPU only CPU (faster-whisper) CPU (Kokoro)

WSL2 為開發環境,GPU 不可用,所有測試以 CPU + mock 模式執行。


環境變數

變數 說明 預設
AUREKA_CONFIG config.toml 路徑 ./config.toml
AUREKA_TEST_MODE 1 跳過模型載入(測試加速)
AUREKA_LOG_LEVEL debug / info / warning info

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

aureka-0.1.1.tar.gz (29.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

aureka-0.1.1-py3-none-any.whl (24.5 kB view details)

Uploaded Python 3

File details

Details for the file aureka-0.1.1.tar.gz.

File metadata

  • Download URL: aureka-0.1.1.tar.gz
  • Upload date:
  • Size: 29.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for aureka-0.1.1.tar.gz
Algorithm Hash digest
SHA256 ea61e23f8234fff1995756951ab593d964a2c2c38ea500f4d427a24f918d823b
MD5 89cd826189adae7462c060cf01c410dc
BLAKE2b-256 fa2e9c2db25f19b959bb46e3f965bacfc4dbd2529900fad3a666512a49331bcb

See more details on using hashes here.

Provenance

The following attestation bundles were made for aureka-0.1.1.tar.gz:

Publisher: publish.yml on fcwu/aureka

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file aureka-0.1.1-py3-none-any.whl.

File metadata

  • Download URL: aureka-0.1.1-py3-none-any.whl
  • Upload date:
  • Size: 24.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for aureka-0.1.1-py3-none-any.whl
Algorithm Hash digest
SHA256 e7d360a8df0cae346857c53ad0fa475dc251c8ab7eb562d8511fb5be11cd744e
MD5 4e273be5eec6e1a7f7dbbf170c401255
BLAKE2b-256 d282855750245df6d63678f6be75daab06dc605d19be57c6ded45e160228010a

See more details on using hashes here.

Provenance

The following attestation bundles were made for aureka-0.1.1-py3-none-any.whl:

Publisher: publish.yml on fcwu/aureka

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page