| |
| |
| |
| |
| |
| |
| |
|
|
| """Extract exactly the six contract boundary fields from validated alignments.""" |
|
|
| from __future__ import annotations |
|
|
| from dataclasses import dataclass |
|
|
| from speech_text_data_aligners.core.errors import AlignmentValidationError |
| from speech_text_data_aligners.core.models import AlignmentRequest, AlignmentResult |
| from speech_text_data_aligners.core.validation import validate_alignment_result |
|
|
| BOUNDARY_COLUMNS = ( |
| "first_word_text", |
| "first_word_start_s", |
| "first_word_end_s", |
| "last_word_text", |
| "last_word_start_s", |
| "last_word_end_s", |
| ) |
|
|
|
|
| @dataclass(frozen=True, slots=True) |
| class BoundaryFields: |
| """Hold original-facing first/last lexical surfaces and acoustic boundaries. |
| |
| Parameters |
| ---------- |
| first_word_text: Original surface of the first lexical word. |
| first_word_start_s: Acoustic start of the first lexical word. |
| first_word_end_s: Acoustic end of the first lexical word. |
| last_word_text: Original surface of the last lexical word. |
| last_word_start_s: Acoustic start of the last lexical word. |
| last_word_end_s: Acoustic end of the last lexical word. |
| |
| """ |
|
|
| first_word_text: str |
| first_word_start_s: float |
| first_word_end_s: float |
| last_word_text: str |
| last_word_start_s: float |
| last_word_end_s: float |
|
|
| def as_dict(self) -> dict[str, str | float]: |
| """Return fields in the exact output-schema insertion order. |
| |
| Returns: |
| A new ordered dictionary with the six contract column names. |
| |
| """ |
| return { |
| "first_word_text": self.first_word_text, |
| "first_word_start_s": self.first_word_start_s, |
| "first_word_end_s": self.first_word_end_s, |
| "last_word_text": self.last_word_text, |
| "last_word_start_s": self.last_word_start_s, |
| "last_word_end_s": self.last_word_end_s, |
| } |
|
|
|
|
| def extract_boundary_fields( |
| request: AlignmentRequest, |
| result: AlignmentResult, |
| ) -> BoundaryFields: |
| """Validate a result and select its first/last lexical alignments. |
| |
| A one-word utterance intentionally uses the same record for both endpoints. |
| |
| Parameters |
| ---------- |
| request: Authoritative request used for semantic validation. |
| result: Candidate complete result. |
| |
| Returns |
| ------- |
| Exactly six backend-neutral boundary values using original word surfaces. |
| |
| Raises |
| ------ |
| AlignmentValidationError: If validation fails or no aligned words exist. |
| TranscriptError: If the authoritative transcript has no lexical words. |
| |
| """ |
| validated = validate_alignment_result(request, result) |
| if not validated.words: |
| raise AlignmentValidationError("cannot extract boundaries from no words") |
| first = validated.words[0] |
| last = validated.words[-1] |
| return BoundaryFields( |
| first_word_text=first.lexical_word.text, |
| first_word_start_s=first.start_s, |
| first_word_end_s=first.end_s, |
| last_word_text=last.lexical_word.text, |
| last_word_start_s=last.start_s, |
| last_word_end_s=last.end_s, |
| ) |
|
|
