Skip to content

armgit / andee

Public Imported from GitHub
0

Andee lives on your phone or Mac and uses any app like you do—no integrations, no root, any model; memory and passwords never leave your device | Andee 住进你的手机或 Mac,像你一样操作任何 App——无需集成、无需 Root、支持任意模型,记忆和密码绝不离开设备。

docs: name the two ASR resource ids this build can use, and rule out 2.0

booboosuic1b54606e443
diff --git a/README.md b/README.md
index 314d125..392ab0a 100644
--- a/README.md
+++ b/README.md
@@ -530,10 +530,16 @@ error code. Neither message points at 开通管理, which is why it is written d
**If the pack you bought is a different SKU, the 识别资源 ID row is the third cause.** It
sits under the Volcengine key (`asr_resource_id`), defaults to `volc.bigasr.sauc.duration`
-(the by-duration pack) and is sent as the `X-Api-Resource-Id` header, so a by-concurrency
-pack (并发版) needs its own value there instead. Get it wrong and the symptom is
-indistinguishable from the two above: the key is fine, the call just fails. Leave that row
-alone unless you bought something other than the duration pack.
+(the **1.0** model's by-duration pack) and is sent as the `X-Api-Resource-Id` header. A
+by-concurrency pack (并发版) of the same model takes `volc.bigasr.sauc.concurrent` there
+instead. Get it wrong and the symptom is indistinguishable from the two above: the key is
+fine, the call just fails.
+
+The two ids that pair with this build are the 1.0 ones, because the request it sends also names
+the model: `model_name` is `bigmodel`, which is 豆包流式语音识别模型 1.0. The 2.0 service
+(豆包流式语音识别 2.0, resource ids `volc.seedasr.sauc.duration` / `volc.seedasr.sauc.concurrent`)
+is a different model, so if the console only offered you 2.0, this row cannot rescue it — open
+the 1.0 service as well. Leave the row alone once it holds one of the two 1.0 values.
> **If you took the CodeFlying route in [Install](#install), both keys are already baked in.**
> ⚙ → **Backend** will show **CodeFlying** as the active tab; LLM, ASR, TTS and the long tail of
diff --git a/README.zh-CN.md b/README.zh-CN.md
index b20339d..5d66099 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -435,9 +435,14 @@ API 只认后一句。刚开通的服务默认是**后付费**,它只能在火
`ASR ws refused: …` 或者一个服务器错误码。这两句提示都不会叫你去开通管理,所以这件事写在这里。
**如果买的包是另一种 SKU,第三种原因在「识别资源 ID」那一行。** 它就在火山 key 下面
-(`asr_resource_id`),默认 `volc.bigasr.sauc.duration`(按时长计费的包),作为
-`X-Api-Resource-Id` 头发出去,所以买的是并发版(按并发计费)就得把这里换成对应的值。填错了,
-症状和上面两种分不出来:key 没问题,就是调不通。没买时长版以外的包,这一行就别动它。
+(`asr_resource_id`),默认 `volc.bigasr.sauc.duration`(**1.0** 模型的时长版),作为
+`X-Api-Resource-Id` 头发出去;同一个模型的并发版则要换成 `volc.bigasr.sauc.concurrent`。填错了,
+症状和上面两种分不出来:key 没问题,就是调不通。
+
+能跟这一版配上的是 1.0 那两个值,因为发出去的请求里也写死了模型:`model_name` 是 `bigmodel`,
+也就是**豆包流式语音识别模型 1.0**。**2.0**(豆包流式语音识别 2.0,资源 id 是
+`volc.seedasr.sauc.duration` / `volc.seedasr.sauc.concurrent`)是另一个模型,所以控制台只给你
+2.0 的话,改这一行救不回来 —— 要另外把 1.0 的服务也开通。这一行已经填了上面两个 1.0 值之一,就别动它。
> **如果你走的是[安装](#安装)里 CodeFlying 那条路,这两把 Key 都已经在安装包里了。**
> ⚙ → **后端**会直接显示 **CodeFlying** 这一 tab 为当前后端;LLM、ASR、TTS 以及海量第三方 API
diff --git a/docs/getting-started.md b/docs/getting-started.md
index 868b80c..946200f 100644
--- a/docs/getting-started.md
+++ b/docs/getting-started.md
@@ -29,7 +29,7 @@ The same order is summarized in the README’s [Provisioning order](../README.md
Open ⚙ on the ball and choose **Backend → On-device**. Enter an OpenAI-compatible endpoint and `llm_api_key`; the local brain cannot run without the brain key. If you installed a CodeFlying package, its **CodeFlying** backend is already configured instead. A voice key is optional for a first typed interaction, but voice input and spoken replies need the selected voice provider configured too.
-For the Volcengine voice provider, a key on its own is not enough. A service that shows as activated in the Volcengine console is **post-paid** by default, and that tier works only from the console's demo page in a browser: the app's own calls are refused even when the key is correct. On the console's **开通管理** (activation management) page, change the service to prepaid/subscription and buy a streaming-ASR resource pack there. Until that is done, voice fails at the very start of the turn — the handshake is refused before any audio is sent — or the connection does open and an error frame carrying an entitlement code comes back after the first request. Neither is a network problem. If you bought a by-concurrency pack rather than the by-duration one, also set **识别资源 ID** (ASR resource ID) to the value that pack uses; it defaults to `volc.bigasr.sauc.duration` and is sent as the `X-Api-Resource-Id` header.
+For the Volcengine voice provider, a key on its own is not enough. A service that shows as activated in the Volcengine console is **post-paid** by default, and that tier works only from the console's demo page in a browser: the app's own calls are refused even when the key is correct. On the console's **开通管理** (activation management) page, change the service to prepaid/subscription and buy a streaming-ASR resource pack there. Until that is done, voice fails at the very start of the turn — the handshake is refused before any audio is sent — or the connection does open and an error frame carrying an entitlement code comes back after the first request. Neither is a network problem. If you bought a by-concurrency pack rather than the by-duration one, also set **识别资源 ID** (ASR resource ID) to `volc.bigasr.sauc.concurrent`; it defaults to `volc.bigasr.sauc.duration`, and both are sent as the `X-Api-Resource-Id` header. Those two are the 1.0 model's ids and the request also names the model (`model_name` is `bigmodel`), so a 2.0 pack (豆包流式语音识别 2.0, `volc.seedasr.*`) will not work here — open the 1.0 service as well.
Now tap the ball, type a simple request such as “What can you see on this screen?”, and send it. The ball should expand into the conversation card, answer, and return to the floating ball when it is done. This confirms the basic loop: the app is running, the brain is reachable, and the overlay is visible. Do not use a payment, messaging, or other consequential task as the first test.
diff --git a/docs/getting-started.zh-CN.md b/docs/getting-started.zh-CN.md
index 493fef0..3234565 100644
--- a/docs/getting-started.zh-CN.md
+++ b/docs/getting-started.zh-CN.md
@@ -29,7 +29,7 @@ APK 装到设备后,按下面四步进行:
点击悬浮球上的 ⚙,进入**后端 → 本机**。填入一个 OpenAI 兼容端点和 `llm_api_key`;没有大脑密钥,本机大脑无法运行。如果安装的是 CodeFlying 包,**CodeFlying** 后端已经配置好。第一次用文字互动时,语音密钥不是必需的;但语音输入和朗读回复需要配置所选语音服务商。
-火山语音服务这边,光有 key 不够。控制台里显示「已开通」的服务默认是**后付费**,而它只能在官网的体验页里用:程序走 API 会被拒绝,key 填对了也一样。要到控制台的**开通管理**页把该服务改成预付 / 预购,并在那里买一个流式语音识别的资源包。在这之前,语音会在这一轮的最开头就失败 —— 握手被直接拒绝,音频还没发出去;或者反过来,连接起来了,但第一次请求之后收到一个带授权错误码的错误帧。两种都不是网络问题。如果买的是并发版的包而不是时长版,还要把**识别资源 ID** 改成那个包对应的值;它默认是 `volc.bigasr.sauc.duration`,会作为 `X-Api-Resource-Id` 头发送。
+火山语音服务这边,光有 key 不够。控制台里显示「已开通」的服务默认是**后付费**,而它只能在官网的体验页里用:程序走 API 会被拒绝,key 填对了也一样。要到控制台的**开通管理**页把该服务改成预付 / 预购,并在那里买一个流式语音识别的资源包。在这之前,语音会在这一轮的最开头就失败 —— 握手被直接拒绝,音频还没发出去;或者反过来,连接起来了,但第一次请求之后收到一个带授权错误码的错误帧。两种都不是网络问题。如果买的是并发版的包而不是时长版,还要把**识别资源 ID** 改成 `volc.bigasr.sauc.concurrent`;它默认是 `volc.bigasr.sauc.duration`,两者都会作为 `X-Api-Resource-Id` 头发送。这两个值属于 1.0 模型,而请求里也写死了模型(`model_name` 是 `bigmodel`),所以 **2.0** 的包(豆包流式语音识别 2.0,`volc.seedasr.*`)在这里用不了 —— 要另外把 1.0 的服务也开通。
现在点击悬浮球,输入一句简单的话,例如“你能看到这个屏幕上的什么?”,然后发送。悬浮球应该展开成对话卡片,返回答案,完成后再缩回悬浮球。这说明基本链路已经通了:应用在运行、大脑可连接、悬浮窗可见。第一次测试不要使用付款、发消息或其它有实际后果的任务。