Voice Conversation AI Assistant
Budget: $30 – $250 USD
I’m looking for a straightforward proof-of-concept that lets someone talk to an AI and immediately hear the answer back—no keyboard, no screen taps, just voice in and voice out.
Core goals
• Continuous, push-to-talk or wake-word listening that sends the captured audio to a speech-to-text engine (Whisper, Google STT, or another option you prefer).
• The transcribed text goes to an LLM (ChatGPT, GPT-4, or a similarly capable model) with a system prompt that keeps replies casual and friendly.
• The returned text is converted to natural-sounding speech with a TTS service (e.g., ElevenLabs, Azure, or open-source).
• The assistant should detect the speaker’s language and respond in the same tongue—English, Spanish, Mandarin and any other language the underlying engines support.
Deliverables
1. A runnable demo (Python script, small desktop app, or lightweight mobile prototype—whatever is fastest to test).
2. Clear setup notes listing APIs, keys and versions used so I can replicate the build on my side.
3. A brief read-me explaining how to swap models or voices.
This is an initial concept, so keep it as lean as possible while still demonstrating smooth conversational flow. If the experience feels natural, we can explore richer features later.
Core goals
• Continuous, push-to-talk or wake-word listening that sends the captured audio to a speech-to-text engine (Whisper, Google STT, or another option you prefer).
• The transcribed text goes to an LLM (ChatGPT, GPT-4, or a similarly capable model) with a system prompt that keeps replies casual and friendly.
• The returned text is converted to natural-sounding speech with a TTS service (e.g., ElevenLabs, Azure, or open-source).
• The assistant should detect the speaker’s language and respond in the same tongue—English, Spanish, Mandarin and any other language the underlying engines support.
Deliverables
1. A runnable demo (Python script, small desktop app, or lightweight mobile prototype—whatever is fastest to test).
2. Clear setup notes listing APIs, keys and versions used so I can replicate the build on my side.
3. A brief read-me explaining how to swap models or voices.
This is an initial concept, so keep it as lean as possible while still demonstrating smooth conversational flow. If the experience feels natural, we can explore richer features later.