46 lines
2.0 KiB
Markdown
46 lines
2.0 KiB
Markdown
---
|
||
title: "Two-Way Voice Conversation"
|
||
type: concept
|
||
tags: [voice, conversation, interaction, AI, real-time]
|
||
sources: [phone-call-notifications]
|
||
last_updated: 2026-04-23
|
||
---
|
||
|
||
## Definition
|
||
|
||
Two-Way Voice Conversation 是一种实时双向语音交互模式——AI Agent 主动拨叫用户,用户接听后可随时提问,Agent 实时响应。与传统的单向播报(TTS)或录音通知不同,真正的双向对话意味着用户可以打断、追问、澄清,而不是被动接收广播式信息。
|
||
|
||
## Why It Matters
|
||
|
||
传统通知(推送、邮件、SMS)的本质是**单向广播**——Agent 发送信息,用户被动接收,无法即时互动。Two-Way Voice Conversation 打破了这一限制:
|
||
|
||
- **即时澄清**:用户可以说"等等,哪封邮件?"或"现在价格多少?",Agent 立即查询并回答
|
||
- **动态深度**:用户可追问"为什么跌了?"引导 Agent 深入解释
|
||
- **决策支持**:电话中即可做出决策("帮我取消"、"确认出席"),无需挂断后额外操作
|
||
|
||
## Contrast with One-Way Notification
|
||
|
||
| Aspect | One-Way Notification | Two-Way Voice Conversation |
|
||
|--------|---------------------|---------------------------|
|
||
| Direction | Agent → User | Agent ↔ User |
|
||
| User Action | Passive receive | Active query |
|
||
| Depth | Fixed message | Dynamic based on questions |
|
||
| Decision Making | Post-call action needed | Can decide on-call |
|
||
| Latency Tolerance | Low (pre-composed) | Higher (live response) |
|
||
|
||
## Design Considerations
|
||
|
||
- **模型速度**:通话场景需要低延迟响应,建议使用快速模型(Haiku 级别)减少等待
|
||
- **上下文管理**:Agent 需要在通话中维护对话上下文,切换话题时自然流畅
|
||
- **通话礼仪**:Agent 应简洁明了,不说废话,直接回答用户问题
|
||
|
||
## Related Concepts
|
||
|
||
- [[Voice Notification Channel]] — 双向语音是电话通知的核心能力
|
||
- [[Voice Interface]] — 语音交互的更宽泛概念
|
||
- [[Telephony Integration]] — 电话集成的技术实现
|
||
|
||
## Sources
|
||
|
||
- [[phone-call-notifications]]
|