lattifai 0.4.0__py3-none-any.whl → 0.4.1__py3-none-any.whl

lattifai/bin/agent.py

@@ -210,7 +210,7 @@ async def _run_youtube_workflow(
     click.echo()
 
     # Import workflow components
-    from lattifai import AsyncLattifAI
+    from lattifai.client import AsyncLattifAI
     from lattifai.workflows import YouTubeSubtitleAgent
     from lattifai.workflows.gemini import GeminiTranscriber
     from lattifai.workflows.youtube import YouTubeDownloader
@@ -309,12 +309,13 @@ if __name__ == '__main__':
 
     asyncio.run(
         _run_youtube_workflow(
+            # url='https://www.youtube.com/watch?v=7nv1snJRCEI',
             url='https://www.youtube.com/watch?v=DQacCB9tDaw',
             lattifai_api_key=os.getenv('LATTIFAI_API_KEY'),
             gemini_api_key=os.getenv('GEMINI_API_KEY', ''),
-            device='cpu',
+            device='mps',
             model_name_or_path='Lattifai/Lattice-1-Alpha',
-            media_format='mp4',
+            media_format='mp3',
             output_format='TextGrid',
             output_dir='~/Downloads/lattifai_youtube',
             max_retries=0,

lattifai/io/parser.py

@@ -0,0 +1,75 @@
+import logging
+import re
+from typing import Optional, Tuple
+
+# 来自于字幕中常见的说话人标记格式
+SPEAKER_PATTERN = re.compile(r'((?:>>|>>|>|>).*?[:：])\s*(.*)')
+
+# Transcriber Output Example:
+# 26:19.919 --> 26:34.921
+# [SPEAKER_01]: 越来越多的科技巨头入...
+SPEAKER_LATTIFAI = re.compile(r'(^\[SPEAKER_.*?\]:)\s*(.*)')
+
+# NISHTHA BHATIA: Hey, everyone.
+# DIETER: Oh, hey, Nishtha.
+# GEMINI: That might
+SPEAKER_PATTERN2 = re.compile(r'^([A-Z]{1,15}(?:\s+[A-Z]{1,15})?)[:：]\s*(.*)$')
+
+
+def parse_speaker_text(line) -> Tuple[Optional[str], str]:
+    line = line.replace('\\N', ' ')
+
+    if ':' not in line and '：' not in line:
+        return None, line
+
+    # 匹配以 >> 开头的行，并去除开头的名字和冒号
+    match = SPEAKER_PATTERN.match(line)
+    if match:
+        return match.group(1).strip(), match.group(2).strip()
+
+    match = SPEAKER_LATTIFAI.match(line)
+    if match:
+        assert len(match.groups()) == 2, match.groups()
+        if not match.group(1):
+            logging.error(f'ParseSub LINE [{line}]')
+        else:
+            return match.group(1).strip(), match.group(2).strip()
+
+    match = SPEAKER_PATTERN2.match(line)
+    if match:
+        assert len(match.groups()) == 2, match.groups()
+        return match.group(1).strip(), match.group(2).strip()
+
+    return None, line
+
+
+if __name__ == '__main__':
+    pattern = re.compile(r'>>\s*(.*?)\s*[:：]\s*(.*)')
+    pattern = re.compile(r'(>>.*?[:：])\s*(.*)')
+
+    test_strings = [
+        '>>Key: Value',
+        '>>  Key with space : Value with space ',
+        '>>  全角键 ： 全角值',
+        '>>Key：Value xxx. >>Key：Value',
+    ]
+
+    for text in test_strings:
+        match = pattern.match(text)
+        if match:
+            print(f"Input: '{text}'")
+            print(f"  Key:   '{match.group(1)}'")
+            print(f"  Value: '{match.group(2)}'")
+            print('-------------')
+
+    # pattern2
+    test_strings2 = ['NISHTHA BHATIA: Hey, everyone.', 'DIETER: Oh, hey, Nishtha.', 'GEMINI: That might']
+    for text in test_strings2:
+        match = SPEAKER_PATTERN2.match(text)
+        if match:
+            print(f"  Input: '{text}'")
+            print(f"Speaker: '{match.group(1)}'")
+            print(f"Content: '{match.group(2)}'")
+            print('-------------')
+        else:
+            raise ValueError(f"No match for: '{text}'")

lattifai/io/reader.py

@@ -4,6 +4,7 @@ from typing import List, Literal, Optional, Union
 
 from lhotse.utils import Pathlike
 
+from .parser import parse_speaker_text
 from .supervision import Supervision
 
 SubtitleFormat = Literal['txt', 'srt', 'vtt', 'ass', 'auto']
@@ -72,9 +73,11 @@ class SubtitleReader(ABCMeta):
 
         supervisions = []
         for event in subs.events:
+            speaker, text = parse_speaker_text(event.text)
             supervisions.append(
                 Supervision(
-                    text=event.text,
+                    text=text,
+                    speaker=speaker,
                     start=event.start / 1000.0 if event.start is not None else None,
                     duration=(event.end - event.start) / 1000.0 if event.end is not None else None,
                 )

lattifai/tokenizer/tokenizer.py

@@ -192,12 +192,15 @@ class LatticeTokenizer:
         for s, supervision in enumerate(supervisions):
             text_len += len(supervision.text)
             if supervision.speaker:
-                speakers.append(supervision.speaker)
                 if sidx < s:
+                    if len(speakers) < len(texts) + 1:
+                        speakers.append(None)
                     text = ' '.join([sup.text for sup in supervisions[sidx:s]])
                     texts.append(text)
                     sidx = s
                     text_len = len(supervision.text)
+                speakers.append(supervision.speaker)
+
             else:
                 if text_len >= 2000 or s == len(supervisions) - 1:
                     if len(speakers) < len(texts) + 1:
@@ -228,7 +231,7 @@ class LatticeTokenizer:
                     remainder = ''
                 # Detect and split special sentence types: e.g., '[APPLAUSE] &gt;&gt; MIRA MURATI:' -> ['[APPLAUSE]', '&gt;&gt; MIRA MURATI:']  # noqa: E501
                 resplit_parts = self._resplit_special_sentence_types(_sentence)
-                if any(resplit_parts[-1].endswith(sp) for sp in [':', '：']):
+                if any(resplit_parts[-1].endswith(sp) for sp in [':', '：', ']']):
                     if s < len(_sentences) - 1:
                         _sentences[s + 1] = resplit_parts[-1] + ' ' + _sentences[s + 1]
                     else:  # last part
@@ -238,6 +241,12 @@ class LatticeTokenizer:
                     processed_sentences.extend(resplit_parts)
             _sentences = processed_sentences
 
+            if not _sentences:
+                if remainder:
+                    _sentences, remainder = [remainder.strip()], ''
+                else:
+                    continue
+
             if any(_sentences[-1].endswith(ep) for ep in END_PUNCTUATION):
                 supervisions.extend(
                     Supervision(text=text, speaker=(_speaker if s == 0 else None)) for s, text in enumerate(_sentences)

{lattifai-0.4.0.dist-info → lattifai-0.4.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lattifai
-Version: 0.4.0
+Version: 0.4.1
 Summary: Lattifai Python SDK: Seamless Integration with Lattifai's Speech and Video AI Services
 Author-email: Lattifai Technologies <tech@lattifai.com>
 Maintainer-email: Lattice <tech@lattifai.com>
@@ -40,7 +40,6 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
-Classifier: Programming Language :: Python :: 3.14
 Classifier: Operating System :: MacOS :: MacOS X
 Classifier: Operating System :: POSIX :: Linux
 Classifier: Operating System :: Microsoft :: Windows
@@ -135,16 +134,15 @@ The library provides two equivalent commands: `lai` (recommended for convenience
 ```bash
 # Align audio with subtitle (using lai command)
 lai align audio.wav subtitle.srt output.srt
-
 # Or use the full command
 lattifai align audio.wav subtitle.srt output.srt
 
-# Process YouTube videos with intelligent workflow
-lai agent --youtube https://www.youtube.com/watch?v=VIDEO_ID
-
 # Download and align YouTube content directly
 lai youtube https://www.youtube.com/watch?v=VIDEO_ID
 
+# Process YouTube videos with intelligent workflow (equivalent to lai youtube)
+lai agent --youtube https://www.youtube.com/watch?v=VIDEO_ID
+
 # Convert subtitle format
 lai subtitle convert input.srt output.vtt
 ```
@@ -261,7 +259,8 @@ Options:
 ```
 
 **When to use `lai agent` vs `lai youtube`**:
-- **Use `lai agent`**: For production workflows, batch processing, advanced error handling, and when you need retry logic
+- Both `lai agent --youtube URL` and `lai youtube URL` provide the same core functionality for downloading and aligning YouTube content
+- **Use `lai agent --youtube`**: For production workflows, batch processing, advanced error handling, and when you need retry logic
 - **Use `lai youtube`**: For quick one-off downloads and alignment with minimal overhead
 
 #### Understanding --split_sentence

{lattifai-0.4.0.dist-info → lattifai-0.4.1.dist-info}/RECORD

@@ -4,25 +4,26 @@ lattifai/client.py,sha256=WewbRnfI8b_rQ8V2JG3y2ojkiyEWylXI7No_4NpbzEI,13282
 lattifai/errors.py,sha256=5i_acoly1g-TLAID8QnhzQshwOXfgLL55mHsdwzlNGA,10814
 lattifai/utils.py,sha256=CzVwNc08u8lm7XavCMJskXHfni0xsZ-EgpcMkRywVm8,4736
 lattifai/bin/__init__.py,sha256=QWbmVUbzqwPQNeOV_g-bOsz53w37v-tbZ3uFrSj-7Mg,90
-lattifai/bin/agent.py,sha256=7e_Q1ND62u-_voKpLkO77WHmmInJYDPPL0hQuKq8LCY,9863
+lattifai/bin/agent.py,sha256=daXsg0PsW4HS53PYw-r6fSSDsoe9XHOIVEzz-djjKto,9935
 lattifai/bin/align.py,sha256=Tpu2OWBXBYUqzBt1Qu6HmkVZ8Bpx6mc0O9KItDimoCc,8377
 lattifai/bin/cli_base.py,sha256=i4KQDyUPJFKae_wjuQeB4N2Vhg9vZ6yc6L9YUJLC18I,507
 lattifai/bin/subtitle.py,sha256=UZMPh71O2X1UwbfZ9VWlhzxkz78viz8KWwoVsDpewK0,6577
 lattifai/io/__init__.py,sha256=mGnQkGm6BClRcogRK3J1nJJfPOk5z4o0sZI2nnyJVOU,1087
 lattifai/io/gemini_reader.py,sha256=WDZA93MSrUAsa5j-ZDXLdPXzEIoREymEy-rMAED_6f4,15152
 lattifai/io/gemini_writer.py,sha256=rlXO9zx6kQhqTi9K9izE69-8S-2GPOIiJHPwZyebpiM,6515
-lattifai/io/reader.py,sha256=_tVQPQ6KBal3Zd7kq1IZ7rpidE0tDmfpk64iU7cGPrg,3199
+lattifai/io/parser.py,sha256=LQHgcEYXaSdhwUo9rP6P_31Z6RMv_BTP1YSKzXji4bk,2386
+lattifai/io/reader.py,sha256=wm4O4ry2ZtBbGtO1FQTrTa1r2_83gUfZpzdh61qvuu0,3328
 lattifai/io/supervision.py,sha256=iBDRiDJ0hddo__SoEZau2cdEIBFnXZNLgSWFjtJd-lM,871
 lattifai/io/utils.py,sha256=4drRwcM1n7AYhdJcF51EZxMTy_Ut_1GKtdWpRhPuVmg,686
 lattifai/io/writer.py,sha256=8n9ZBuXuVOCFwzr1hqrnXpZ-fARTsepebwjKgRuueWE,3872
 lattifai/tokenizer/__init__.py,sha256=y-FyfO7tLga9b46pkCC6jdSBKOFZS-jFfHcqUieGEyU,120
 lattifai/tokenizer/phonemizer.py,sha256=SfRi1KIMpmaao6OVmR1h_I_3QU-vrE6D5bh72Afg5XM,1759
-lattifai/tokenizer/tokenizer.py,sha256=AIliPRfBscqzyc2z-P638rOmyF38FbliHkE1NQzrwsE,18903
+lattifai/tokenizer/tokenizer.py,sha256=m6dDvvDmaVFtQd0aikMUViRZNzztXM0OyFs-vlToT9o,19191
 lattifai/workers/__init__.py,sha256=s6YfkIq4FDIAzY9sPjRpXnJfszj2repqnMTqydRM5Zw,83
 lattifai/workers/lattice1_alpha.py,sha256=1lCq0-bgWMXvYslAbCTFgHC0p6UWPto1y0wkTw9WrmQ,10177
-lattifai-0.4.0.dist-info/licenses/LICENSE,sha256=LNuoH5jpXXNKgjQ3XLwztFq8D3O7kZI-LSg81o4ym2M,1065
-lattifai-0.4.0.dist-info/METADATA,sha256=VLBwud4AA78kRomOnJ0_dD6tLruRp13DQk3xephebmE,26597
-lattifai-0.4.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lattifai-0.4.0.dist-info/entry_points.txt,sha256=fCgo8-LKA_9C7_jmEGsZPJko0woXHtEh0iRbpO7PYzI,69
-lattifai-0.4.0.dist-info/top_level.txt,sha256=-OVWZ68YYFcTN13ARkLasp2OUappe9wEVq-CKes7jM4,17
-lattifai-0.4.0.dist-info/RECORD,,
+lattifai-0.4.1.dist-info/licenses/LICENSE,sha256=LNuoH5jpXXNKgjQ3XLwztFq8D3O7kZI-LSg81o4ym2M,1065
+lattifai-0.4.1.dist-info/METADATA,sha256=MLSUT7RUUOUd0lD_NA3GIR-AjBBkwjZV-HLiNK5k534,26719
+lattifai-0.4.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lattifai-0.4.1.dist-info/entry_points.txt,sha256=fCgo8-LKA_9C7_jmEGsZPJko0woXHtEh0iRbpO7PYzI,69
+lattifai-0.4.1.dist-info/top_level.txt,sha256=-OVWZ68YYFcTN13ARkLasp2OUappe9wEVq-CKes7jM4,17
+lattifai-0.4.1.dist-info/RECORD,,