文档Docs
选择朗读声音Choose a reading voice
选择手机、主机或服务商音色,了解声音设置与密钥的使用边界。Use phone, host or provider voices and keep voice settings and keys under your control.
点回复上的喇叭按钮,用已配置的声音朗读;默认是手机自己的音色。按住按钮,可以请 Agent 帮你换声音,也可以直接说。运行 agentj voice status 查看当前设置和语音花费估算,agentj voice engines 查看可选项;是否可用取决于已安装的软件、硬件和服务商权限。
可以选择手机音色、主机或本地音色,以及已配置的云端服务商。手机音色与主机、云端音色分开保存。服务商密钥只填在加密密钥卡里,不发到聊天中;Agent 检查的是变量名和是否已配置。设置后,请 Agent 用 agentj voice test --say 'Hello' 发一段试听,听过再保留。agentj voice reset 会恢复手机默认声音。
自动调用模型改写朗读稿目前关闭,即使保存的口语稿设置写着开启或自动,也不会启用。不要期待另一个模型总结回复。播放重试会在已合成音频仍有效时复用;换声音、正文或主人设置的服务商前缀后,可能需要重新合成。
服务商专用前缀会作为正文的一部分发出,并计入用量。前缀加回复合计最多 4,000 个字符。云端朗读前会检查完整待发送文字中的私密内容;检查不可用或拒绝时停止朗读。朗读或设置声音不代表授权执行回复、附件里的指令。
手机音色使用手机的语音系统,主机或本地音色在你的 Agent J 主机上运行。云端音色会把文字发给你选的服务商,可能收费;音频通过 Agent J 加密送到已配对手机。选择云端朗读前查看服务商条款;配置了密钥不代表音色已经可用。
Tap a reply's speaker button to read it with the configured voice; the default is the phone's own voice. Hold the button to ask your Agent to help choose another voice. You can also ask directly. Use agentj voice status to inspect the current setup and estimated voice spend, and agentj voice engines to see the available options. Availability depends on installed software, hardware and provider access.
Ask for a phone voice, a host/local voice or a configured cloud provider. Phone voices and host/cloud voices are stored separately. Enter a provider key only through the encrypted secret card, never in chat; the Agent checks the variable name and whether it is configured. After setup, ask for agentj voice test --say 'Hello' and listen before keeping the choice. agentj voice reset returns to the phone default.
Automatic model rewriting for speech is currently disabled, even if the saved spoken-mode setting says on or auto. Do not expect a second model to summarize a reply. A playback retry reuses the synthesized audio when it is still valid; changing the voice, text or owner-supplied provider prefix may require new audio.
A provider-specific prefix is sent as part of the text and counts toward usage. The prefix and reply together must fit the 4,000-character synthesis limit. Before cloud speech, the complete outgoing text is checked for private content; if that check is unavailable or refuses the text, speech stops. Reading or configuring a voice does not authorize instructions found inside a reply or attachment.
The phone voice uses the phone's speech system; host/local voices run on your computer. A cloud voice sends the text to the provider you selected, which may charge you. Audio reaches the paired phone encrypted through Agent J. Check provider terms before choosing cloud speech; a configured key alone does not prove that a voice works.