e11c3abb9e
- Added DashScope ASR service implementation for real-time streaming. - Updated ASR provider logic to support DashScope alongside existing providers. - Enhanced runtime metadata resolution to include DashScope as a valid ASR provider. - Modified configuration files and documentation to reflect the addition of DashScope. - Introduced tests to validate DashScope integration and ASR service behavior. - Refactored ASR service factory to accommodate new provider options and modes.
30 lines
966 B
Markdown
30 lines
966 B
Markdown
# 语音识别
|
||
|
||
语音识别（ASR）负责将用户音频实时转写为文本，供对话引擎理解。
|
||
|
||
## 模式
|
||
|
||
- `offline`：引擎本地缓冲音频后触发识别（适用于 OpenAI-compatible / SiliconFlow）。
|
||
- `streaming`：音频分片实时发送到服务端，服务端持续返回转写事件（适用于 DashScope Realtime ASR）。
|
||
|
||
## 配置项
|
||
|
||
| 配置项 | 说明 |
|
||
|---|---|
|
||
| ASR 引擎 | 选择语音识别服务提供商 |
|
||
| 模型 | 识别模型名称 |
|
||
| 语言 | 中文/英文/多语言 |
|
||
| 热词 | 提升特定词汇识别准确率 |
|
||
| 标点与规范化 | 是否自动补全标点、文本规范化 |
|
||
|
||
## 建议
|
||
|
||
- 客服场景建议开启热词并维护业务词表
|
||
- 多语言场景建议按会话入口显式指定语言
|
||
- 对延迟敏感场景优先选择流式识别模型
|
||
- 当前支持提供商：`openai_compatible`、`siliconflow`、`dashscope`、`buffered`（回退）
|
||
|
||
## 相关文档
|
||
|
||
- [语音配置总览](voices.md)
|