update: complete the doc for alignment pipeline

ml/lab/align-pipeline.md

@@ -1,3 +1,4 @@
 1. prepare `1.mp3`, `1.txt`, `1.group` in `./data`
-2. `whisperAlignment/alignWithGroup`
-3. `mmsAlignment
+2. Run `whisperAlignment/alignWithGroup.py`
+3. Run `mmsAlignment/splitSong.py`
+4. Run `mmsAlignment/alignWithMMS.py`

ml/lab/mmsAlignment/alignWithMMS.py

@@ -164,4 +164,4 @@ def align(audio_file: str, lrc_file: str, output_ttml: str, segments_dir: str =
     ttml_generator.save(output_ttml)
     
 if __name__ == "__main__":
-    align("./data/1.flac", "./data/1.lrc", "./data/output.ttml", "./temp/lines")
+    align("./data/1.mp3", "./data/1.lrc", "./data/output.ttml", "./temp/lines")

ml/lab/mmsAlignment/splitSong.py

@@ -52,7 +52,7 @@ def split_audio_by_lrc(audio_file, lrc_data):
 
 if __name__ == "__main__":
     lrc_file = "./data/1.lrc"  # LRC文件路径
-    audio_file = "./data/1.flac"  # 音频文件路径
+    audio_file = "./data/1.mp3"  # 音频文件路径
     
     lrc_data = parse_lrc(lrc_file)
     split_audio_by_lrc(audio_file, lrc_data)

ml/lab/whisperAlignment/align2srt.py

@@ -3,4 +3,13 @@ import stable_whisper
 def align2srt(lyrics, audio_path, output_path):
     model = stable_whisper.load_model('large-v3')
     result = model.align(audio_path, lyrics, language="Chinese", regroup=False)
-    result.to_srt_vtt(output_path, segment_level=False)
+    result.to_srt_vtt(output_path, segment_level=False)
+
+AUDIO_FILE = './data/1.mp3'
+LYRICS_FILE = './data/1.txt'
+OUTPUT_FILE = './data/1.srt'
+
+if __name__ == "__main__":
+    with open(LYRICS_FILE, 'r') as f:
+        lyrics_content = f.read()
+        align2srt(lyrics_content, AUDIO_FILE, OUTPUT_FILE)