fix: extract the language code wrapped in a markup (#494)

Fixes #492

Signed-off-by: Frost Ming <me@frostming.com>

pyproject.toml

@@ -25,7 +25,7 @@ dependencies = [
     "google-generativeai",
     "numexpr>=2.8.6",
     "dashscope>=1.10.0",
-    "tetos>=0.1.1",
+    "tetos>=0.2.1",
 ]
 license = {text = "MIT"}
 dynamic = ["version", "optional-dependencies"]

requirements.txt

@@ -78,7 +78,7 @@ socksio==1.0.0
 soupsieve==2.5
 sqlalchemy==2.0.25
 tenacity==8.2.3
-tetos==0.1.1
+tetos==0.2.1
 tqdm==4.66.1
 typing-extensions==4.9.0
 typing-inspect==0.9.0

xiaogpt/config.py

@@ -76,7 +76,9 @@ class Config:
     start_conversation: str = "开始持续对话"
     end_conversation: str = "结束持续对话"
     stream: bool = False
-    tts: Literal["mi", "edge", "azure", "openai", "baidu", "google", "volc"] = "mi"
+    tts: Literal[
+        "mi", "edge", "azure", "openai", "baidu", "google", "volc", "minimax"
+    ] = "mi"
     tts_options: dict[str, Any] = field(default_factory=dict)
     gpt_options: dict[str, Any] = field(default_factory=dict)
     bing_cookie_path: str = ""

xiaogpt/tts/tetos.py

@@ -4,7 +4,6 @@
 from pathlib import Path
 
 from miservice import MiNAService
-from tetos.base import Speaker
 
 from xiaogpt.config import Config
 from xiaogpt.tts.base import AudioFileTTS
@@ -14,39 +13,15 @@ class TetosTTS(AudioFileTTS):
     def __init__(
         self, mina_service: MiNAService, device_id: str, config: Config
     ) -> None:
-        super().__init__(mina_service, device_id, config)
-        self.speaker = self._get_speaker()
-
-    def _get_speaker(self) -> Speaker:
-        from tetos.azure import AzureSpeaker
-        from tetos.baidu import BaiduSpeaker
-        from tetos.edge import EdgeSpeaker
-        from tetos.google import GoogleSpeaker
-        from tetos.openai import OpenAISpeaker
-        from tetos.volc import VolcSpeaker
+        from tetos import get_speaker
 
-        options = self.config.tts_options
-        allowed_speakers: list[str] = []
-        for speaker in (
-            OpenAISpeaker,
-            EdgeSpeaker,
-            AzureSpeaker,
-            VolcSpeaker,
-            GoogleSpeaker,
-            BaiduSpeaker,
-        ):
-            if (name := speaker.__name__[:-7].lower()) == self.config.tts:
-                try:
-                    return speaker(**options)
-                except TypeError as e:
-                    raise ValueError(
-                        f"{e}. Please add them via `tts_options` config"
-                    ) from e
-            else:
-                allowed_speakers.append(name)
-        raise ValueError(
-            f"Unsupported TTS: {self.config.tts}, allowed: {','.join(allowed_speakers)}"
-        )
+        super().__init__(mina_service, device_id, config)
+        assert config.tts and config.tts != "mi"
+        speaker_cls = get_speaker(config.tts)
+        try:
+            self.speaker = speaker_cls(**config.tts_options)
+        except TypeError as e:
+            raise ValueError(f"{e}. Please add them via `tts_options` config") from e
 
     async def make_audio_file(self, lang: str, text: str) -> tuple[Path, float]:
         output_file = tempfile.NamedTemporaryFile(

xiaogpt/xiaogpt.py

@@ -418,9 +418,15 @@ async def speak(self, text_stream: AsyncIterator[str]) -> None:
             # It is not a legal language code, discard it
             lang, first_chunk = "", text
 
+        lang = (
+            matches[0]
+            if (matches := re.findall(r"([a-z]{2}-[A-Z]{2})", lang))
+            else "zh-CN"
+        )
+
         async def gen():  # reconstruct the generator
             yield first_chunk
             async for text in text_stream:
                 yield text
 
-        await self.tts.synthesize(lang or "zh-CN", gen())
+        await self.tts.synthesize(lang, gen())