lattifai 1.2.1__py3-none-any.whl → 1.3.0__py3-none-any.whl

lattifai-1.3.0.dist-info/RECORD

@@ -0,0 +1,57 @@
+lattifai/_init.py,sha256=ZpiYRpP7kPh6XpQK6nX1zKKvtQE3Xr0T2Ed5S4wDiwU,609
+lattifai/audio2.py,sha256=JpMvwB0e09hLoffji8zOSlgrIhqUUq_zq0ubT_QMspc,15317
+lattifai/client.py,sha256=d6WpodzseoSv6aA8jkj4hwiMOz3irSkD15i1hr2JUsQ,18182
+lattifai/errors.py,sha256=aPTMhTaeQjY0KMifnm-Kpo9MQPsoYsEanZB2f0AA6qQ,11911
+lattifai/logging.py,sha256=MbUEeOUFlF92pA9v532DiPPWKl03S7UHCJ6Z652cf0w,2860
+lattifai/mixin.py,sha256=_d-kzu2w5Y4mZTFMEumh30baWzKsQ9n_IS432woi5vQ,26295
+lattifai/types.py,sha256=JK7KVaZhX89BiKPm4okY0DWLHY1S8aj-YiZXoVH1akw,667
+lattifai/utils.py,sha256=5LeunAN0OQ1jWoKMIThpXSEOxFYD2dCRTdsglosodUU,7963
+lattifai/alignment/__init__.py,sha256=aOyC1P5DqESNLpDh6Gu6LyUZAVMba-IKI7Ugz7v9G4w,344
+lattifai/alignment/lattice1_aligner.py,sha256=MvBpPnhfF8NYtH2ANhQszKNRQROUiSYrBz3aN1bdT3U,11120
+lattifai/alignment/lattice1_worker.py,sha256=Z7hxaS-nucNsUmrphbD8tgBBYPkJOgQb-85nFON94_I,13041
+lattifai/alignment/phonemizer.py,sha256=fbhN2DOl39lW4nQWKzyUUTMUabg7v61lB1kj8SKK-Sw,1761
+lattifai/alignment/punctuation.py,sha256=qLcvuXhBzoEa6bznWZiAB5TAxR6eLr_ZV-PnnCY90UA,1218
+lattifai/alignment/segmenter.py,sha256=JTbBYEXn8hkFwy0tITORy7nKoUPiNYLfi3w1DJNeHZ0,6303
+lattifai/alignment/text_align.py,sha256=sF-6Tsf863BhJcii3joeNa6Auv-7l3SiOhh9j8oPGME,14935
+lattifai/alignment/tokenizer.py,sha256=OIpMGHg1rJ7n97zncDMPpXy32uGOSt1yXiNO4sO6eP0,18839
+lattifai/cli/__init__.py,sha256=PdqoCTqRSFSWrqL3FjBTa5VzJy_e6Rq0OzyT7YkyHpc,541
+lattifai/cli/alignment.py,sha256=rqg6wU2vf6RJ058yWVoXft_UJfOCrEpmE-ye5fhTphg,6129
+lattifai/cli/caption.py,sha256=jkMme73sJ16dkVpRh7O6qjbr14SUeBif00vCTBn7ed0,10339
+lattifai/cli/diarization.py,sha256=GTd2vnTm6cJN6Q3mFP-ShY9bZBl1_zKzWFu-4HHcMzk,4075
+lattifai/cli/transcribe.py,sha256=vZIV0TCbZG_IL2F_Mg49cCGSCBinOOFAtROajVTpNWE,7853
+lattifai/cli/youtube.py,sha256=FJwDl48-cuacP1sdPvX19vdszXdT7EoOZgGYzJpoLeM,6360
+lattifai/config/__init__.py,sha256=nJUVk03JRj4rujoEmkCkQ8akZF7kqIj7ci3XphU9uVA,1249
+lattifai/config/alignment.py,sha256=3JUtgHBueIK_lH9PgeBPjuHGL4VvDEYVs9fvylir6bc,5392
+lattifai/config/caption.py,sha256=OMLsW8QKDWM6A3G5V3Gf-9bgB3D1PC5gO8LiiNNeOwM,7195
+lattifai/config/client.py,sha256=qqHKFPV4iEjVHCDOuGx7kj-tYFtgZZAszOQRFsNFbO8,2359
+lattifai/config/diarization.py,sha256=cIkwCfsYqfMns3i6tKWcwBBBkdnhhmB_Eo0TuOPCw9o,2484
+lattifai/config/event.py,sha256=P-_2yOzSATZSXz-ctlWeJQGOKCbNLFnWLBvUZ8JclyA,3845
+lattifai/config/media.py,sha256=nxvgC7zeLsthCARPPUbnK2eMJY8R1d-1XgiAsy8kroA,15568
+lattifai/config/transcription.py,sha256=V0WtZ_p-WsBienRbGyd-zLdX6F_XRsDWGlba_qzwet0,4115
+lattifai/data/__init__.py,sha256=hdUhvlUjPgb3_Hd_cJ30f2oXHBMZRGzaSafd64b3vYA,168
+lattifai/data/caption.py,sha256=MVuZiQ47Lr3A1afFqGkqFzpWjPakmsQusQ86t210Y2Y,7800
+lattifai/diarization/__init__.py,sha256=-ZZ_a5hIQgnlHIOehCTtmVmWOWC2H6eOhSs4AcVtRtk,1782
+lattifai/diarization/lattifai.py,sha256=tCnFL6ywITqeKR8YoCsYvyJxNoIwoC6GsnI9zkXNB-Q,3128
+lattifai/event/__init__.py,sha256=PPAWzrkRK8YgWhG6CtIUkb7nH8svd9_zGOhxjz0_dcM,2448
+lattifai/event/lattifai.py,sha256=QJqUxJsIWryVVoud_qE8af6zoJ89ZyPgHDvQp4OzXg0,5826
+lattifai/transcription/__init__.py,sha256=vMHciyCEPKhhfM3KjMCeDqnyxU1oghF8g5o5SvpnT_4,2669
+lattifai/transcription/base.py,sha256=ywRjIGg6emTx1v8PCSPyHcdugR6PvdTl10H64Iu1iqs,4617
+lattifai/transcription/gemini.py,sha256=p6uZlhPQuzzUsj226Jk-INOt7NF5g4TIN6yEn1ZwrBI,18030
+lattifai/transcription/lattifai.py,sha256=DA7QSN-a_yIZq79Nc_f6lf8_VWW4qqhyXfoZ1Um-31M,3451
+lattifai/transcription/prompts/README.md,sha256=X49KWSQVdjWxxWUp4R2w3ZqKrAOi6_kDNHh1hMaQ4PE,694
+lattifai/transcription/prompts/__init__.py,sha256=G9b42COaCYv3sPPNkHsGDLOMBuVGKt4mXGYal_BYtYQ,1351
+lattifai/transcription/prompts/gemini/README.md,sha256=rt7f7yDGtaobKBo95LG3u56mqa3ABOXQd0UVgJYtYuo,781
+lattifai/transcription/prompts/gemini/transcription_gem.txt,sha256=cljzZ--BDgnnKzqVCakr-fTp2Xk38UOsUquvruNX-LU,4600
+lattifai/workflow/__init__.py,sha256=INpQgc9gZ2Fp-aTHcpR3TEHGtEtPzjOB8T7-jLzVM0E,1547
+lattifai/workflow/agents.py,sha256=yEOnxnhcTvr1iOhCorNvp8B76P6nQsLRXJCu_rCYFfM,38
+lattifai/workflow/base.py,sha256=8QoVIBZwJfr5mppJbtUFafHv5QR9lL-XrULjTWD0oBg,6257
+lattifai/workflow/file_manager.py,sha256=yc29Vb7JNUMJ9rwM_YjkAHfDInl8HMVAl9A7z7XiIOU,32974
+lattifai/youtube/__init__.py,sha256=_uO3KCx-t6I-JaYFpcYLYpvkbmEOOni3xBqGEbExg68,1587
+lattifai/youtube/client.py,sha256=VU8FC1N7YYpbc4LeJNAsahNAI1R7e3_7Yjmb1rz7tyI,52878
+lattifai/youtube/types.py,sha256=80RgBmvM4tRbxqyNv9GU6hr9vPp_yhKrK0RJ_vG2h4E,472
+lattifai-1.3.0.dist-info/licenses/LICENSE,sha256=xGMLmdFJy6Jkz3Hd0znyQLmcxC93FSZB5isKnEDMoQQ,1066
+lattifai-1.3.0.dist-info/METADATA,sha256=WMgLRzKiJv_Zn1aoxPjWofNZRE4tSjTWxa16zWNYVTk,23008
+lattifai-1.3.0.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
+lattifai-1.3.0.dist-info/entry_points.txt,sha256=MfoqXNjXrhD7VMApHgaHmAECTcGVUMUiR0uqnTg7Ads,502
+lattifai-1.3.0.dist-info/top_level.txt,sha256=tHSoXF26r-IGfbIP_JoYATqbmf14h5NrnNJGH4j5reI,9
+lattifai-1.3.0.dist-info/RECORD,,

{lattifai-1.2.1.dist-info → lattifai-1.3.0.dist-info}/entry_points.txt

@@ -1,11 +1,10 @@
 [console_scripts]
 lai-align = lattifai.cli.alignment:main
-lai-app-install = lattifai.cli.app_installer:main
 lai-diarize = lattifai.cli.diarization:main
-lai-server = lattifai.cli.server:main
 lai-transcribe = lattifai.cli.transcribe:main
 lai-youtube = lattifai.cli.youtube:main
 laicap-convert = lattifai.cli.caption:main_convert
+laicap-diff = lattifai.cli.caption:main_diff
 laicap-normalize = lattifai.cli.caption:main_normalize
 laicap-shift = lattifai.cli.caption:main_shift
 

lattifai/__init__.py

@@ -1,88 +0,0 @@
-import os
-import warnings
-from importlib.metadata import version
-
-# Suppress SWIG deprecation warnings before any imports
-warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*SwigPy.*")
-
-# Suppress PyTorch transformer nested tensor warning
-warnings.filterwarnings("ignore", category=UserWarning, message=".*enable_nested_tensor.*")
-
-# Disable tokenizers parallelism warning
-os.environ["TOKENIZERS_PARALLELISM"] = "false"
-
-# Re-export I/O classes
-from .caption import Caption
-
-# Re-export client classes
-from .client import LattifAI
-
-# Re-export config classes
-from .config import (
-    AUDIO_FORMATS,
-    MEDIA_FORMATS,
-    VIDEO_FORMATS,
-    AlignmentConfig,
-    CaptionConfig,
-    ClientConfig,
-    DiarizationConfig,
-    MediaConfig,
-)
-from .errors import (
-    AlignmentError,
-    APIError,
-    AudioFormatError,
-    AudioLoadError,
-    AudioProcessingError,
-    CaptionParseError,
-    CaptionProcessingError,
-    ConfigurationError,
-    DependencyError,
-    LatticeDecodingError,
-    LatticeEncodingError,
-    LattifAIError,
-    ModelLoadError,
-)
-from .logging import get_logger, set_log_level, setup_logger
-
-try:
-    __version__ = version("lattifai")
-except Exception:
-    __version__ = "0.1.0"  # fallback version
-
-
-__all__ = [
-    # Client classes
-    "LattifAI",
-    # Config classes
-    "AlignmentConfig",
-    "ClientConfig",
-    "CaptionConfig",
-    "DiarizationConfig",
-    "MediaConfig",
-    "AUDIO_FORMATS",
-    "VIDEO_FORMATS",
-    "MEDIA_FORMATS",
-    # Error classes
-    "LattifAIError",
-    "AudioProcessingError",
-    "AudioLoadError",
-    "AudioFormatError",
-    "CaptionProcessingError",
-    "CaptionParseError",
-    "AlignmentError",
-    "LatticeEncodingError",
-    "LatticeDecodingError",
-    "ModelLoadError",
-    "DependencyError",
-    "APIError",
-    "ConfigurationError",
-    # Logging
-    "setup_logger",
-    "get_logger",
-    "set_log_level",
-    # I/O
-    "Caption",
-    # Version
-    "__version__",
-]

lattifai/alignment/sentence_splitter.py

@@ -1,219 +0,0 @@
-import re
-from typing import List, Optional
-
-from lattifai.caption import Supervision
-from lattifai.utils import _resolve_model_path
-
-END_PUNCTUATION = '.!?"]。！？"】'
-
-
-class SentenceSplitter:
-    """Lazy-initialized sentence splitter using wtpsplit."""
-
-    def __init__(self, device: str = "cpu", model_hub: Optional[str] = None, lazy_init: bool = True):
-        """Initialize sentence splitter with lazy loading.
-
-        Args:
-            device: Device to run the model on (cpu, cuda, mps)
-            model_hub: Model hub to use (None for huggingface, "modelscope" for modelscope)
-        """
-        self.device = device
-        self.model_hub = model_hub
-        if lazy_init:
-            self._splitter = None
-        else:
-            self._init_splitter()
-
-    def _init_splitter(self):
-        """Initialize the sentence splitter model on first use."""
-        if self._splitter is not None:
-            return
-
-        import onnxruntime as ort
-        from wtpsplit import SaT
-
-        providers = []
-        device = self.device
-        if device.startswith("cuda") and ort.get_all_providers().count("CUDAExecutionProvider") > 0:
-            providers.append("CUDAExecutionProvider")
-        elif device.startswith("mps") and ort.get_all_providers().count("MPSExecutionProvider") > 0:
-            providers.append("MPSExecutionProvider")
-
-        if self.model_hub == "modelscope":
-            downloaded_path = _resolve_model_path("LattifAI/OmniTokenizer", model_hub="modelscope")
-            sat = SaT(
-                f"{downloaded_path}/sat-3l-sm",
-                tokenizer_name_or_path=f"{downloaded_path}/xlm-roberta-base",
-                ort_providers=providers + ["CPUExecutionProvider"],
-            )
-        else:
-            sat_path = _resolve_model_path("segment-any-text/sat-3l-sm", model_hub="huggingface")
-            sat = SaT(
-                sat_path,
-                tokenizer_name_or_path="facebookAI/xlm-roberta-base",
-                hub_prefix="segment-any-text",
-                ort_providers=providers + ["CPUExecutionProvider"],
-            )
-        self._splitter = sat
-
-    @staticmethod
-    def _resplit_special_sentence_types(sentence: str) -> List[str]:
-        """
-        Re-split special sentence types.
-
-        Examples:
-        '[APPLAUSE] >> MIRA MURATI:' -> ['[APPLAUSE]', '>> MIRA MURATI:']
-        '[MUSIC] >> SPEAKER:' -> ['[MUSIC]', '>> SPEAKER:']
-
-        Special handling patterns:
-        1. Separate special marks at the beginning (e.g., [APPLAUSE], [MUSIC], etc.) from subsequent speaker marks
-        2. Use speaker marks (>> or other separators) as split points
-
-        Args:
-            sentence: Input sentence string
-
-        Returns:
-            List of re-split sentences. If no special marks are found, returns the original sentence in a list
-        """
-        # Detect special mark patterns: [SOMETHING] >> SPEAKER:
-        # or other forms like [SOMETHING] SPEAKER:
-
-        # Pattern 1: [mark] HTML-encoded separator speaker:
-        pattern1 = r"^(\[[^\]]+\])\s+(>>|>>)\s+(.+)$"
-        match1 = re.match(pattern1, sentence.strip())
-        if match1:
-            special_mark = match1.group(1)
-            separator = match1.group(2)
-            speaker_part = match1.group(3)
-            return [special_mark, f"{separator} {speaker_part}"]
-
-        # Pattern 2: [mark] speaker:
-        pattern2 = r"^(\[[^\]]+\])\s+([^:]+:)(.*)$"
-        match2 = re.match(pattern2, sentence.strip())
-        if match2:
-            special_mark = match2.group(1)
-            speaker_label = match2.group(2)
-            remaining = match2.group(3).strip()
-            if remaining:
-                return [special_mark, f"{speaker_label} {remaining}"]
-            else:
-                return [special_mark, speaker_label]
-
-        # If no special pattern matches, return the original sentence
-        return [sentence]
-
-    def split_sentences(self, supervisions: List[Supervision], strip_whitespace=True) -> List[Supervision]:
-        """Split supervisions into sentences using the sentence splitter.
-
-        Careful about speaker changes.
-
-        Args:
-            supervisions: List of Supervision objects to split
-            strip_whitespace: Whether to strip whitespace from split sentences
-
-        Returns:
-            List of Supervision objects with split sentences
-        """
-        self._init_splitter()
-
-        texts, speakers = [], []
-        text_len, sidx = 0, 0
-
-        def flush_segment(end_idx: int, speaker: Optional[str] = None):
-            """Flush accumulated text from sidx to end_idx with given speaker."""
-            nonlocal text_len, sidx
-            if sidx <= end_idx:
-                if len(speakers) < len(texts) + 1:
-                    speakers.append(speaker)
-                text = " ".join(sup.text for sup in supervisions[sidx : end_idx + 1])
-                texts.append(text)
-                sidx = end_idx + 1
-                text_len = 0
-
-        for s, supervision in enumerate(supervisions):
-            text_len += len(supervision.text)
-            is_last = s == len(supervisions) - 1
-
-            if supervision.speaker:
-                # Flush previous segment without speaker (if any)
-                if sidx < s:
-                    flush_segment(s - 1, None)
-                    text_len = len(supervision.text)
-
-                # Check if we should flush this speaker's segment now
-                next_has_speaker = not is_last and supervisions[s + 1].speaker
-                if is_last or next_has_speaker:
-                    flush_segment(s, supervision.speaker)
-                else:
-                    speakers.append(supervision.speaker)
-
-            elif text_len >= 2000 or is_last:
-                flush_segment(s, None)
-
-        assert len(speakers) == len(texts), f"len(speakers)={len(speakers)} != len(texts)={len(texts)}"
-        sentences = self._splitter.split(texts, threshold=0.15, strip_whitespace=strip_whitespace, batch_size=8)
-
-        supervisions, remainder = [], ""
-        for k, (_speaker, _sentences) in enumerate(zip(speakers, sentences)):
-            # Prepend remainder from previous iteration to the first sentence
-            if _sentences and remainder:
-                _sentences[0] = remainder + _sentences[0]
-                remainder = ""
-
-            if not _sentences:
-                continue
-
-            # Process and re-split special sentence types
-            processed_sentences = []
-            for s, _sentence in enumerate(_sentences):
-                if remainder:
-                    _sentence = remainder + _sentence
-                    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 s < len(_sentences) - 1:
-                        _sentences[s + 1] = resplit_parts[-1] + " " + _sentences[s + 1]
-                    else:  # last part
-                        remainder = resplit_parts[-1] + " "
-                    processed_sentences.extend(resplit_parts[:-1])
-                else:
-                    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)
-                )
-                _speaker = None  # reset speaker after use
-            else:
-                supervisions.extend(
-                    Supervision(text=text, speaker=(_speaker if s == 0 else None))
-                    for s, text in enumerate(_sentences[:-1])
-                )
-                remainder = _sentences[-1] + " " + remainder
-                if k < len(speakers) - 1 and speakers[k + 1] is not None:  # next speaker is set
-                    supervisions.append(
-                        Supervision(text=remainder.strip(), speaker=_speaker if len(_sentences) == 1 else None)
-                    )
-                    remainder = ""
-                elif len(_sentences) == 1:
-                    if k == len(speakers) - 1:
-                        pass  # keep _speaker for the last supervision
-                    else:
-                        assert speakers[k + 1] is None
-                        speakers[k + 1] = _speaker
-                else:
-                    assert len(_sentences) > 1
-                    _speaker = None  # reset speaker if sentence not ended
-
-        if remainder.strip():
-            supervisions.append(Supervision(text=remainder.strip(), speaker=_speaker))
-
-        return supervisions

lattifai/caption/__init__.py

@@ -1,20 +0,0 @@
-from typing import List, Optional
-
-from lhotse.utils import Pathlike
-
-from ..config.caption import InputCaptionFormat
-from .caption import Caption
-from .gemini_reader import GeminiReader, GeminiSegment
-from .gemini_writer import GeminiWriter
-from .supervision import Supervision
-from .text_parser import normalize_text
-
-__all__ = [
-    "Caption",
-    "Supervision",
-    "GeminiReader",
-    "GeminiWriter",
-    "GeminiSegment",
-    "normalize_text",
-    "InputCaptionFormat",
-]