OmniVoice：将 Alexa 智能音箱接入任意大语言模型的开源方案

OmniVoice is an open-source Alexa skill that bridges Amazon Echo and other smart speakers with any OpenAI-compatible large language models (LLMs like OpenAI, Gemini, Groq). It allows users to enjoy open, context-aware voice interactions without coding, server maintenance, or AWS accounts, using free Alexa-Hosted Skills for deployment. Key benefits include zero friction setup, session memory, global support, and time zone awareness.

Current smart speakers (Amazon Alexa, Google Assistant) are limited by preset commands and closed ecosystems, lacking open dialogue. Meanwhile, powerful LLMs (ChatGPT, Claude, Gemini) lack convenient voice interaction entry points. OmniVoice addresses this by combining the two, enabling smart speakers to use advanced LLMs for natural conversations.

OmniVoice is a fully open-source Alexa skill with three core design principles: zero friction (no coding/server/AWS account),通用性强 (works with any OpenAI-compatible LLM), and high optimization (meets Alexa's 8-second timeout). It uses Alexa-Hosted Skills for free backend hosting, allowing users to deploy in minutes.

Data Flow: User voice → Alexa → AWS Lambda (Python backend) → LLM API → response → Alexa voice播报 (under 8s).

Key Features:

• Open text capture: Uses AMAZON.SearchQuery slot and dialogue prefixes to capture natural questions.
• Low latency: Progressive response ("processing...") to avoid timeout.
• Privacy: Sensitive info (API keys) stored in .env (excluded from Git).
• Session memory: Maintains 10 rounds of history (context-aware, within 24KB limit).
• Global support: Localized for major English regions.
• Time zone awareness: Injects current time/date into system prompts for time-sensitive queries.

Deployment Steps:

1. Create OmniVoice skill in Alexa Developer Console (Custom + Alexa-Hosted Python).
2. Import code from GitHub repo.
3. Copy .env.example to .env and add API keys (OpenAI, Groq, etc.).
4. Deploy and build the model, then test with "Alexa, open Omni Voice".

Model Recommendations:

• Preferred: Google Gemini 2.5 Flash (via OpenRouter, 1-1.5s response).
• Fastest: Groq's Llama models (0.2-0.4s response).
• Avoid: DeepSeek-R1 or congested APIs (risk of timeout).

Reason: Alexa has strict 8-second timeout, so low-latency models are critical.

Skill Exits Suddenly: Caused by unrecognized phrases. Fixed by using fallbackIntentSensitivity (HIGH) and routing to FallbackIntent for polite prompts. "Skill Response Error": Most often due to invalid API keys or insufficient balance. Check CloudWatch logs for .env variable correctness (no spaces around equals).

OmniVoice is MIT-licensed and hosted on GitHub, welcoming contributions (bug fixes, new features, docs). It represents a trend of integrating LLMs into daily hardware, breaking closed ecosystems. For users: revives idle smart speakers; for devs: learning case for Alexa skills, Lambda, LLM integration. Future possibilities: multi-modal models, edge inference for faster responses.