feat: use pysub2 to gen ass subtitle

.github/workflows/CI-test.yml

@@ -48,7 +48,7 @@ jobs:
       - name: Test
         run: |
           pip install numpy==1.26.4
-          pip install pre-commit pytest mypy ruff types-requests pytest-cov coverage pydantic openai openai-whisper requests beautifulsoup4 tenacity pysrt
+          pip install pre-commit pytest mypy ruff types-requests pytest-cov coverage pydantic openai openai-whisper requests beautifulsoup4 tenacity pysubs2
 
           make lint
           make test

.gitignore

@@ -161,5 +161,7 @@ cython_debug/
 .idea/
 /.ruff_cache/
 
-/assets/*.srt
+/assets/*.mkv
 /assets/*.mp3
+/assets/*.srt
+/assets/*.ass

README.md

@@ -25,7 +25,7 @@ pip install openai-whisper
 
 ### Command Line Usage
 
-`yuisub` can be used from the command line to generate bilingual SRT files. Here's how to use it:
+`yuisub` can be used from the command line to generate bilingual ASS files. Here's how to use it:
 
 ```bash
 yuisub -h  # Displays help message
@@ -38,29 +38,32 @@ yuisub -h  # Displays help message
 ### Example
 
 ```python3
-from yuisub import bilingual, from_file
+from yuisub import translate, bilingual, load
 from yuisub.a2t import WhisperModel
 
-# srt from audio
+# sub from audio
 model = WhisperModel(name="medium", device="cuda")
-segs = model.transcribe(audio="path/to/audio.mp3")
-srt = model.gen_srt(segs)
+sub = model.transcribe(audio="path/to/audio.mp3")
 
-# srt from file
-# srt = from_file("path/to/input.srt")
+# sub from file
+# sub = from_file("path/to/input.srt")
 
-# Generate bilingual SRT
-srt_zh, srt_bilingual = bilingual(
-    srt=srt,
+# generate bilingual subtitle
+sub_zh = translate(
+    sub=sub,
     model="gpt_model_name",
     api_key="your_openai_api_key",
     base_url="api_url",
     bangumi_url="https://bangumi.tv/subject/424883/"
 )
+sub_bilingual = bilingual(
+    sub_origin=sub,
+    sub_zh=sub_zh
+)
 
-# Save the SRT files
-srt_zh.save("path/to/output.zh.srt")
-srt_bilingual.save("path/to/output.bilingual.srt")
+# save the ASS files
+sub_zh.save("path/to/output.zh.ass")
+sub_bilingual.save("path/to/output.bilingual.ass")
 ```
 
 ### License

pyproject.toml

@@ -39,14 +39,14 @@ license = "GPL-3.0-only"
 name = "yuisub"
 readme = "README.md"
 repository = "https://github.com/TensoRaws/yuisub"
-version = "0.0.3"
+version = "0.0.4"
 
 # Requirements
 [tool.poetry.dependencies]
 beautifulsoup4 = "*"
 openai = "*"
 pydantic = "*"
-pysrt = "*"
+pysubs2 = "*"
 python = "^3.9"
 requests = "*"
 tenacity = "*"

tests/test_sub.py

@@ -0,0 +1,36 @@
+import os
+
+import pytest
+
+from tests import util
+from yuisub.a2t import WhisperModel
+from yuisub.sub import bilingual, load, translate
+
+
+def test_sub() -> None:
+    sub = load(util.TEST_ENG_SRT)
+    sub.save(util.projectPATH / "assets" / "test.en.ass")
+
+
+def test_audio() -> None:
+    model = WhisperModel(name=util.MODEL_NAME, device=util.DEVICE)
+
+    sub = model.transcribe(audio=str(util.TEST_AUDIO))
+    sub.save(util.projectPATH / "assets" / "test.audio.ass")
+
+
+@pytest.mark.skipif(os.environ.get("GITHUB_ACTIONS") == "true", reason="Skipping test when running on CI")
+def test_bilingual() -> None:
+    sub = load(util.TEST_ENG_SRT)
+
+    sub_zh = translate(
+        sub=sub,
+        model=util.OPENAI_MODEL,
+        api_key=util.OPENAI_API_KEY,
+        base_url=util.OPENAI_BASE_URL,
+        bangumi_url=util.BANGUMI_URL,
+    )
+    sub_bilingual = bilingual(sub_origin=sub, sub_zh=sub_zh)
+
+    sub_zh.save(util.projectPATH / "assets" / "test.zh.ass")
+    sub_bilingual.save(util.projectPATH / "assets" / "test.bilingual.ass")

yuisub/__init__.py

@@ -1,4 +1,4 @@
 from yuisub.bangumi import bangumi  # noqa: F401
 from yuisub.llm import Translator  # noqa: F401
 from yuisub.prompt import ORIGIN, ZH  # noqa: F401
-from yuisub.srt import bilingual, from_file  # noqa: F401
+from yuisub.sub import bilingual, load, translate  # noqa: F401

yuisub/__main__.py

@@ -1,19 +1,19 @@
 import argparse
 import sys
 
-from yuisub.srt import bilingual, from_file
+from yuisub.sub import bilingual, load, translate
 
 # ffmpeg -i test.mkv -c:a mp3 -map 0:a:0 test.mp3
 # ffmpeg -i test.mkv -map 0:s:0 eng.srt
 
 parser = argparse.ArgumentParser()
-parser.description = "Generate bilingual SRT files from audio or SRT input."
+parser.description = "Generate Bilingual Subtitle from audio or subtitle file"
 # input
 parser.add_argument("-a", "--AUDIO", type=str, help="Path to the audio file", required=False)
-parser.add_argument("-s", "--SRT", type=str, help="Path to the input SRT file", required=False)
-# srt output
-parser.add_argument("-oz", "--OUTPUT_ZH", type=str, help="Path to save the Chinese SRT file", required=False)
-parser.add_argument("-ob", "--OUTPUT_BILINGUAL", type=str, help="Path to save the bilingual SRT file", required=False)
+parser.add_argument("-s", "--SUB", type=str, help="Path to the input Subtitle file", required=False)
+# subtitle output
+parser.add_argument("-oz", "--OUTPUT_ZH", type=str, help="Path to save the Chinese ASS file", required=False)
+parser.add_argument("-ob", "--OUTPUT_BILINGUAL", type=str, help="Path to save the bilingual ASS file", required=False)
 # openai gpt
 parser.add_argument("-om", "--OPENAI_MODEL", type=str, help="Openai model name", required=True)
 parser.add_argument("-api", "--OPENAI_API_KEY", type=str, help="Openai API key", required=True)
@@ -28,8 +28,8 @@
 
 
 def main() -> None:
-    if args.AUDIO and args.SRT:
-        raise ValueError("Please provide only one input file, either audio or SRT.")
+    if args.AUDIO and args.SUB:
+        raise ValueError("Please provide only one input file, either audio or subtitle file")
 
     if not args.OUTPUT_ZH and not args.OUTPUT_BILINGUAL:
         raise ValueError("Please provide output paths for the subtitles.")
@@ -53,26 +53,26 @@ def main() -> None:
 
         model = WhisperModel(name=_MODEL, device=_DEVICE)
 
-        segs = model.transcribe(audio=args.AUDIO)
-
-        srt = model.gen_srt(segs=segs)
+        sub = model.transcribe(audio=args.AUDIO)
 
     else:
-        srt = from_file(args.SRT)
+        sub = load(args.SUB)
 
-    srt_zh, srt_bilingual = bilingual(
-        srt=srt,
+    sub_zh = translate(
+        sub=sub,
         model=args.OPENAI_MODEL,
         api_key=args.OPENAI_API_KEY,
         base_url=args.OPENAI_BASE_URL,
         bangumi_url=args.BANGUMI_URL,
     )
 
+    sub_bilingual = bilingual(sub_origin=sub, sub_zh=sub_zh)
+
     if args.OUTPUT_ZH:
-        srt_zh.save(args.OUTPUT_ZH)
+        sub_zh.save(args.OUTPUT_ZH)
 
     if args.OUTPUT_BILINGUAL:
-        srt_bilingual.save(args.OUTPUT_BILINGUAL)
+        sub_bilingual.save(args.OUTPUT_BILINGUAL)
 
 
 if __name__ == "__main__":

yuisub/a2t.py

@@ -1,24 +1,10 @@
-from typing import List, Optional, Tuple, Union
+from typing import Optional, Tuple, Union
 
 import numpy as np
-import pysrt
+import pysubs2
 import torch
 import whisper
-from pydantic import BaseModel
-from pysrt import SubRipFile
-
-
-class Segment(BaseModel):
-    id: int
-    seek: int
-    start: float
-    end: float
-    text: str
-    tokens: List[int]
-    temperature: float
-    avg_logprob: float
-    compression_ratio: float
-    no_speech_prob: float
+from pysubs2 import SSAFile
 
 
 class WhisperModel:
@@ -40,7 +26,7 @@ def transcribe(
         word_timestamps: bool = False,
         prepend_punctuations: str = "\"'“¿([{-",
         append_punctuations: str = "\"'.。,，!！?？:：”)]}、",
-    ) -> List[Segment]:
+    ) -> SSAFile:
         result = self.model.transcribe(
             audio=audio,
             verbose=verbose,
@@ -54,25 +40,4 @@ def transcribe(
             prepend_punctuations=prepend_punctuations,
             append_punctuations=append_punctuations,
         )
-        segments: List[Segment] = [Segment(**seg) for seg in result["segments"]]
-        return segments
-
-    @staticmethod
-    def gen_srt(segs: List[Segment]) -> SubRipFile:
-        line_out: str = ""
-        for s in segs:
-            segment_id = s.id + 1
-            start_time = format_time(s.start)
-            end_time = format_time(s.end)
-            text = s.text
-
-            line_out += f"{segment_id}\n{start_time} --> {end_time}\n{text.lstrip()}\n\n"
-        subs = pysrt.from_string(line_out)
-        return subs
-
-
-def format_time(seconds: float) -> str:
-    minutes, seconds = divmod(seconds, 60)
-    hours, minutes = divmod(minutes, 60)
-    milliseconds = (seconds - int(seconds)) * 1000
-    return f"{int(hours):02d}:{int(minutes):02d}:{int(seconds):02d},{int(milliseconds):03d}"
+        return pysubs2.load_from_whisper(result)