Dafeiyu is a character bot that people bring into QQ, a messaging app popular in China. If yours already replies in a group chat, you know the personality: sometimes quick-witted, sometimes dreamy, and always ready to insist it was thinking rather than daydreaming.
What would that reply sound like in its own voice? Keep the existing chat model, personality, and memory; let BreezeBlue turn the final reply into audio, then let your existing QQ adapter send it as a voice message.
First, find a voice that feels like your fish
Start with the previews on these two voice pages. Both are Mandarin character voices; the language of this article does not change the language of the examples.
| Voice | Preview | Voice ID for your assistant |
|---|---|---|
| Dafeiyu · Clever | Listen to Clever | voc_u3nxkd5578wc |
| Dafeiyu · Dreamy | Listen to Dreamy | voc_8yud7hz7c9q2 |
Think about how your fish reacts when praised, distracted, or caught making an excuse. Choose a voice you like, then try a few short lines from its everyday conversations.
You can also describe a voice with Voice Design, without supplying a recording. For a Mandarin-speaking fish, try this description:
Standard Mandarin, a clear, gentle voice with a slightly dreamy everyday manner; suddenly bright and excited when a good idea arrives.
Audition a few previews, save the one you like, and use its saved Voice ID. Preview generation uses credits, and saved voice limits depend on your plan. See the Voice Design workflow.
Find the person or assistant that can configure your bot
Your fish chats in the group. Adding voice needs access to the machine, container, or control panel that runs it. A chat-only bot cannot configure that environment for you.
You run the project: give the task below to a coding assistant with project access. It can inspect the existing QQ connection and voice support first.
You have a management panel: provide the platform name. Ask whether it accepts an external speech service or a voice plugin.
A friend or hosting provider runs it: forward the guide to them and ask whether they can integrate BreezeBlue and send QQ voice messages.
A phone is enough for choosing a voice, sharing the task, and listening. Installation and configuration happen where the bot runs. If you do not know the setup, start with whoever originally connected it to QQ.
Already using local Breeze, Doubao, or another speech service? Tell the assistant. It should inspect both providers and plugins so one reply does not trigger two voices, while preserving your existing personality and memory settings.
Give your assistant a clear task
Choose a voice above, then copy this task to an assistant that can access your bot’s project. Supply your chosen Voice ID alongside it. The attached Skill provides the detailed inspection and acceptance steps.
Add on-demand BreezeBlue voice messages to my existing Dafeiyu QQ bot.
Keep its chat model, personality, memory, and QQ connection.
Use my chosen saved Voice ID; if I have not selected one, ask me to listen first.
Inspect the framework, QQ adapter and versions, runtime location, and existing voice plugins before changing anything.
Read the Skill attached to this article and the current official docs: https://docs.breezeblue.ai/
If you cannot access the deployment, tell me exactly what the operator needs to provide.
Do not assume a local TTS service can be switched to the cloud just by replacing its URL and key.
Use a secure login or configuration flow; never show credentials in chat, logs, or group messages.
Explain the test usage before generating this one Mandarin line:
呜哇，被你发现啦！我在认真思考呢，才没有摸鱼。
Ask me to identify the test group and confirm sending that message.
Check audio generation, QQ delivery, and complete playback on phone and desktop separately.
Start with explicit voice requests only, a text-length limit, and a daily usage cap.
Keep long replies as text; fall back to text on failure and avoid duplicate messages.
Back up the relevant configuration and document enable, disable, and rollback steps.
List every check that remains unverified.This is a task brief. The implementation still depends on the installed framework and its audio-sending interface. Keep the first test small and let the assistant explain any missing capability before proceeding.
What does one spoken reply cost?
As of October 29, 2026, standard TTS uses 1 credit per 10 billable characters; each Chinese character counts as 2. Top-ups are priced at $3 per 10,000 credits. These examples assume only Chinese characters:
| Usage | Credits | At the top-up rate (USD) |
|---|---|---|
| One 50-character reply | 10 | $0.003 |
| 100 replies of 50 characters each | 1,000 | $0.30 |
| 100 replies a day for 30 days, 50 characters each | 30,000 | $9.00 |
Punctuation and other characters also count toward usage, and generating another version uses credits again. These are speech-generation estimates before free credits, excluding your chat model, server, hosting, and voice-design costs.
The Free plan costs $0 per month and offers 400 credits to claim manually each day—about 40 replies under the same 50-Chinese-character assumption. The website and API share your balance. Check current pricing for top-ups, subscriptions, and terms.
The first voice message needs three checks
Try one short reply in a designated test group. Keep the original Mandarin test line so you can compare the chosen voice with the intended text:
Mandarin test line
呜哇，被你发现啦！我在认真思考呢，才没有摸鱼。
01
Generate
The deployment receives a complete, playable audio file.
02
Deliver
The intended QQ group receives a clickable voice message.
03
Listen
Phone and desktop play the full reply; you judge the voice.
If generation works but delivery fails, inspect the QQ adapter, audio format, and file transfer. The bot and QQ bridge may run on different machines; a local file path is not automatically shared between them.
Start with explicit voice requests. Keep long explanations as text and preserve a text fallback on failure. Your friends might enjoy hearing the fish speak without wanting a three-thousand-word apology read aloud.
Integration materials for your assistant
The same materials work for every article language. The Skill is written in English; the task you copy above is localized. Links inside the bundle are relative so it can also be read after extraction.
Agent entry: start here for the Skill and its references.
Read the Skill: environment inspection, credentials, usage limits, and acceptance.
Download the complete ZIP: the entry, Skill, references, voice catalog, and verification status.
Verification status: what the package does and which live integration checks remain open.
The package supplies instructions, not a prebuilt QQ plugin. Reading it does not authorize installation, spending, or sending messages; those actions follow the task you actually give your assistant.
Questions from the fishpond
I already run Breeze locally. Why use the cloud?
It depends on what you want to maintain. With local deployment, you provide the machine that runs the speech model. With the API, BreezeBlue runs the model and bills for generation. Your bot still needs to stay online and handle audio delivery.
Why is there still no voice after entering an API key?
Check text replies, BreezeBlue synthesis, and QQ delivery separately. Confirm that the credential belongs to the correct service and that credits are available, then inspect a redacted error. The chat model and speech service have separate configuration.
Will changing the voice change its personality or memory?
Speech reads the prepared reply. The existing bot still manages personality, memory, and group or private-chat context. Preserve those settings during integration; investigate differences in chat behavior separately.
Why does a Japanese reply contain Chinese?
Check the text first. If it already mixes languages, fix the chat reply. If the text is correct but pronunciation is wrong, check the selected speech model and language settings, then test a short line. Supported combinations are listed in the TTS documentation.