feat: 增加双语字幕

decode.py

@@ -32,6 +32,7 @@ class Segment:
     start: float
     duration: float
     text: str = ""
+    cn_text: str = ""
 
     @property
     def end(self):
@@ -44,6 +45,8 @@ class Segment:
         s = s.replace(".", ",")
         s += "\n"
         s += self.text
+        s += "\n"
+        s += self.cn_text
         return s
 
 
@@ -124,12 +127,13 @@ def decode(
 
         for seg, stream in zip(segments, streams):
             en_text = stream.result.text.strip()
-            cn_text = _llm_translator.translate(en_text)
-            seg.text = en_text +"\n"+cn_text
+            seg.text = en_text
             if len(seg.text) == 0:
                 logging.info("Skip empty segment")
                 continue
 
+            seg.cn_text = _llm_translator.translate(en_text)
+
             if len(all_text) == 0:
                 all_text.append(seg.text)
             elif len(all_text[-1][0].encode()) == 1 and len(seg.text[0].encode()) == 1:
@@ -171,7 +175,7 @@ class LLMTranslator:
     def translate(self, src_text: str) -> str:
         translated = self._model.generate(**self._tokenizer(src_text, return_tensors="pt", padding=True))
         res = [self._tokenizer.decode(t, skip_special_tokens=True) for t in translated]
-        return res
+        return "".join(str(itemText) for itemText in res)
 
 
 _llm_translator = LLMTranslator()

requirements.txt

@@ -1,3 +1,6 @@
 #https://huggingface.co/csukuangfj/sherpa-onnx-wheels/resolve/main/cpu/1.10.29/sherpa_onnx-1.10.29-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
 sherpa-onnx>=1.10.35
 ffmpeg-python
+transformers
+sentencepiece
+torch