Repository navigation
2026.4.1
Adds speech synthesis via Qwen3-TTS and tightens how Qwen3 model families are handled across API, CLI, and docs.
Features & API
New OpenAI-compatible endpoint: POST /v1/audio/speech for TTS using Qwen3-TTS models.
Dependencies: qwen-tts, soundfile (see pyproject.toml).
Tests for audio synthesis and request handling on the new route.
CLI & UX
CLI blocks plain text-generation commands on TTS models and points users to the right flows.
anna-speak entry point moved to anna.cli.qwen3_tts_speak:main; README updated for qwen3_tts and anna-speak.
Refactor
API/CLI now route through qwen3_5_text_engine and qwen3_tts_engine for clearer model-family handling.
Removed deprecated speak.py and old model-family inspection tied to it.
Clearer API errors describing which model families support chat vs text completions.
Multimodal handling aligned with the new model structure.
接入 Qwen3-TTS 语音能力,并在 API、CLI 与文档层面统一 Qwen3 系列模型的分流方式。
功能与接口
新增与 OpenAI 风格一致的 POST /v1/audio/speech,用于 Qwen3-TTS 语音合成。
在 pyproject.toml 中增加 qwen-tts、soundfile 等依赖。
为新接口补充测试,覆盖合成与路由请求行为。
CLI 与文档
对 TTS 模型禁止误用纯文本生成类命令,并引导使用正确入口。
anna-speak 入口改为 anna.cli.qwen3_tts_speak:main;README 中补充 qwen3_tts 与 anna-speak 的用法说明。
重构
API 与 CLI 通过 qwen3_5_text_engine、qwen3_tts_engine 区分文本与 TTS 模型族,逻辑更清晰。
删除已弃用的 speak.py 及与之相关的旧模型族探测逻辑。
API 错误信息标明哪些模型族适用于 chat / 文本补全等能力。
多模态处理与新模型结构对齐。