From 428b93f6636a78a7cd7f9c8ce255f265e34dd2a5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marcin=20W=C4=85troba?= <markowanga@gmail.com>
Date: Fri, 29 Apr 2022 08:39:14 +0200
Subject: [PATCH] Add voicelab pipeline stages
---
experiment/pos_processing/__init__.py | 0
.../asr_spacy_token_pos_processing_task.py | 24 ++++++
...nscript_spacy_token_pos_processing_task.py | 24 ++++++
.../spacy_pos_wer_processing_task.py | 85 +++++++++++++++++++
.../spacy_token_pos_processing_task.py | 58 +++++++++++++
5 files changed, 191 insertions(+)
create mode 100644 experiment/pos_processing/__init__.py
create mode 100644 experiment/pos_processing/asr_spacy_token_pos_processing_task.py
create mode 100644 experiment/pos_processing/gold_transcript_spacy_token_pos_processing_task.py
create mode 100644 experiment/pos_processing/spacy_pos_wer_processing_task.py
create mode 100644 experiment/pos_processing/spacy_token_pos_processing_task.py
diff --git a/experiment/pos_processing/__init__.py b/experiment/pos_processing/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/experiment/pos_processing/asr_spacy_token_pos_processing_task.py b/experiment/pos_processing/asr_spacy_token_pos_processing_task.py
new file mode 100644
index 0000000..0abf7b1
--- /dev/null
+++ b/experiment/pos_processing/asr_spacy_token_pos_processing_task.py
@@ -0,0 +1,24 @@
+from dataclasses import dataclass
+from typing import Any
+
+from experiment.pos_processing.spacy_token_pos_processing_task import SpacyTokenPosProcessingTask
+
+
+@dataclass
+class WordSpan:
+ text: str
+ index_start: int
+ index_end: int
+
+
+class AsrSpacyTokenPosProcessingTask(SpacyTokenPosProcessingTask):
+ _spacy_property_name: str
+ _nlp: Any
+ _input_property_name: str
+
+ def __init__(self, task_name: str, spacy_property_name: str, require_update: bool,
+ input_property_name: str):
+ super().__init__(task_name, spacy_property_name, require_update, input_property_name)
+
+ def get_transcript_to_process(self, property_value: Any) -> str:
+ return property_value['full_text']
diff --git a/experiment/pos_processing/gold_transcript_spacy_token_pos_processing_task.py b/experiment/pos_processing/gold_transcript_spacy_token_pos_processing_task.py
new file mode 100644
index 0000000..870a3ac
--- /dev/null
+++ b/experiment/pos_processing/gold_transcript_spacy_token_pos_processing_task.py
@@ -0,0 +1,24 @@
+from dataclasses import dataclass
+from typing import Any
+
+from experiment.pos_processing.spacy_token_pos_processing_task import SpacyTokenPosProcessingTask
+
+
+@dataclass
+class WordSpan:
+ text: str
+ index_start: int
+ index_end: int
+
+
+class GoldTranscriptSpacyTokenPosProcessingTask(SpacyTokenPosProcessingTask):
+ _spacy_property_name: str
+ _nlp: Any
+ _input_property_name: str
+
+ def __init__(self, task_name: str, spacy_property_name: str, require_update: bool,
+ input_property_name: str):
+ super().__init__(task_name, spacy_property_name, require_update, input_property_name)
+
+ def get_transcript_to_process(self, property_value: Any) -> str:
+ return ' '.join([it['word'] for it in property_value])
diff --git a/experiment/pos_processing/spacy_pos_wer_processing_task.py b/experiment/pos_processing/spacy_pos_wer_processing_task.py
new file mode 100644
index 0000000..23d4f57
--- /dev/null
+++ b/experiment/pos_processing/spacy_pos_wer_processing_task.py
@@ -0,0 +1,85 @@
+from typing import Any, List, Dict
+
+from sziszapangma.core.alignment.alignment_classic_calculator import AlignmentClassicCalculator
+from sziszapangma.core.alignment.alignment_step import AlignmentStep
+from sziszapangma.core.wer.wer_calculator import WerCalculator
+from sziszapangma.integration.mapper.alignment_step_mapper import AlignmentStepMapper
+from sziszapangma.integration.repository.experiment_repository import ExperimentRepository
+from sziszapangma.integration.task.processing_task import ProcessingTask
+from sziszapangma.model.model import Word
+from sziszapangma.model.relation_manager import RelationManager
+
+_CLASSIC_WER = "classic_wer"
+
+
+class SpacyPosWerProcessingTask(ProcessingTask):
+ _gold_transcript_pos_property_name: str
+ _asr_pos_property_name: str
+ _pos_alignment_wer: str
+ _pos_metrics_wer: str
+ _alignment_classic_calculator: AlignmentClassicCalculator
+ _wer_calculator: WerCalculator
+
+ def __init__(
+ self,
+ task_name: str,
+ gold_transcript_pos_property_name: str,
+ require_update: bool,
+ asr_pos_property_name: str,
+ pos_alignment_wer: str,
+ pos_metrics_wer: str
+ ):
+ super().__init__(task_name, require_update)
+ self._gold_transcript_pos_property_name = gold_transcript_pos_property_name
+ self._asr_pos_property_name = asr_pos_property_name
+ self._pos_alignment_wer = pos_alignment_wer
+ self._pos_metrics_wer = pos_metrics_wer
+ self._alignment_classic_calculator = AlignmentClassicCalculator()
+ self._wer_calculator = WerCalculator()
+
+ def run_single_process(
+ self,
+ record_id: str,
+ experiment_repository: ExperimentRepository,
+ relation_manager: RelationManager,
+ ):
+ gold_transcript_pos = [
+ Word(id=it['id'], type='Word', text=it['pos'])
+ for it in
+ experiment_repository.get_property_for_key(
+ record_id, self._gold_transcript_pos_property_name)
+ ]
+ asr_transcript_pos = [
+ Word(id=it['id'], type='Word', text=it['pos'])
+ for it in
+ experiment_repository.get_property_for_key(record_id, self._asr_pos_property_name)
+ ]
+ if gold_transcript_pos is not None and asr_transcript_pos is not None:
+ alignment_steps = self._get_alignment(gold_transcript_pos, asr_transcript_pos)
+ experiment_repository.update_property_for_key(
+ record_id,
+ self._pos_alignment_wer,
+ [AlignmentStepMapper.to_json_dict(it) for it in alignment_steps],
+ )
+ experiment_repository.update_property_for_key(
+ record_id, self._pos_metrics_wer, self.calculate_metrics(alignment_steps)
+ )
+
+ def _get_alignment(
+ self, gold_transcript: List[Word], asr_transcript: List[Word]
+ ) -> List[AlignmentStep]:
+ return self._alignment_classic_calculator.calculate_alignment(
+ reference=gold_transcript, hypothesis=asr_transcript
+ )
+
+ def calculate_metrics(self, alignment_steps: List[AlignmentStep]) -> Dict[str, Any]:
+ """Calculate all metrics for data sample."""
+ metrics = dict()
+ metrics[_CLASSIC_WER] = self._wer_calculator.calculate_wer(alignment_steps)
+ return metrics
+
+ def skip_for_record(self, record_id: str, experiment_repository: ExperimentRepository) -> bool:
+ return (
+ experiment_repository.get_property_for_key(record_id, self._pos_metrics_wer)
+ is not None
+ )
diff --git a/experiment/pos_processing/spacy_token_pos_processing_task.py b/experiment/pos_processing/spacy_token_pos_processing_task.py
new file mode 100644
index 0000000..2c8bb0b
--- /dev/null
+++ b/experiment/pos_processing/spacy_token_pos_processing_task.py
@@ -0,0 +1,58 @@
+import uuid
+from abc import abstractmethod, ABC
+from dataclasses import dataclass
+from typing import Any, Dict
+
+import spacy
+from spacy.tokens import Token
+
+from sziszapangma.integration.repository.experiment_repository import ExperimentRepository
+from sziszapangma.integration.task.processing_task import ProcessingTask
+from sziszapangma.model.relation_manager import RelationManager
+
+
+@dataclass
+class WordSpan:
+ text: str
+ index_start: int
+ index_end: int
+
+
+class SpacyTokenPosProcessingTask(ProcessingTask, ABC):
+ _spacy_model_name: str
+ _nlp: Any
+ _input_property_name: str
+ _spacy_property_name: str
+
+ def __init__(
+ self,
+ task_name: str,
+ spacy_property_name: str,
+ require_update: bool,
+ input_property_name: str
+ ):
+ super().__init__(task_name, require_update)
+ self._spacy_property_name = spacy_property_name
+ self._nlp = spacy.load("pl_core_news_lg")
+ self._input_property_name = input_property_name
+
+ def run_single_process(self, record_id: str, experiment_repository: ExperimentRepository,
+ relation_manager: RelationManager):
+ test_property = experiment_repository.get_property_for_key(
+ record_id, self._input_property_name)
+ document = self._nlp(self.get_transcript_to_process(test_property))
+
+ spacy_result = [self.token_to_result_dict(token) for token in document]
+ experiment_repository.update_property_for_key(record_id, self._spacy_property_name,
+ spacy_result)
+
+ @staticmethod
+ def token_to_result_dict(token: Token) -> Dict[str, str]:
+ return {'id': str(uuid.uuid4()), 'word': token.text, 'pos': token.pos_}
+
+ @abstractmethod
+ def get_transcript_to_process(self, property_value: Any) -> str:
+ pass
+
+ def skip_for_record(self, record_id: str, experiment_repository: ExperimentRepository) -> bool:
+ return experiment_repository.property_exists(record_id, self._spacy_property_name)
--
GitLab