cjm-transcript-decomp-core 0.0.10__tar.gz → 0.0.20__tar.gz

{cjm_transcript_decomp_core-0.0.10/cjm_transcript_decomp_core.egg-info → cjm_transcript_decomp_core-0.0.20}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cjm-transcript-decomp-core
-Version: 0.0.10
+Version: 0.0.20
 Summary: A frontend-agnostic core for the transcript decomposition workflow — composes isolated capability workers (forced alignment, VAD, graph storage) into a headless pipeline that decomposes transcription run manifests into a VAD-aligned context-graph spine with traceable provenance, and a CLI as its first driver.
 Author-email: "Christian J. Mills" <9126128+cj-mills@users.noreply.github.com>
 License: Apache-2.0
@@ -14,8 +14,9 @@ Requires-Dist: cjm_substrate>=0.0.63
 Requires-Dist: pyyaml
 Requires-Dist: cjm_context_graph_primitives>=0.0.15
 Requires-Dist: cjm_capability_primitives>=0.0.11
-Requires-Dist: cjm-context-graph-layer>=0.0.11
-Requires-Dist: cjm-transcript-graph-schema>=0.0.7
+Requires-Dist: cjm-context-graph-layer>=0.0.13
+Requires-Dist: cjm-transcript-graph-schema>=0.0.11
+Requires-Dist: cjm-transcription-core>=0.0.21
 Dynamic: license-file
 
 # cjm-transcript-decomp-core
@@ -34,6 +35,8 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - **`cjm_transcript_decomp_core.launch`** — The shared launch surface every decomp shell drives through: the argument
 - **`cjm_transcript_decomp_core.models`** — Lean data shapes for the transcript-decomposition pipeline: in-core mirrors of the forced-alignment / VAD / text DTOs (no FastHTML deps), run configuration, the committed graph-segment carrier, and the decomposition run manifest (proto-bundle).
 - **`cjm_transcript_decomp_core.pipeline`** — The headless decomposition pipeline (stage 5: decomp is an EXTENDER). Load a transcription run manifest, verify the transcription-emitted graph root exists (the graph begins at transcription), then per source per pipeline-segment run VAD + per-transcriber forced alignment, build one aligned segment per VAD chunk with per-transcriber text variants, and attach the fine spine under the existing AudioSegment nodes via the layer's idempotent extend_graph — with HITL approval seams between alignment, commit, and the next source.
+- **`cjm_transcript_decomp_core.respine`** — Chunk RESPINE — re-derive ONE coarse chunk of a LIVE spine from a landed external transcript (work item 7a5e9c84; design ruling 0b4d5cfa, amendment 4a7ec4f8).
+- **`cjm_transcript_decomp_core.retire`** — Spine RETIREMENT + COMPACTION — safe removal of superseded decomposition spines (ruling a7617bd4, item eaefebd2).
 - **`cjm_transcript_decomp_core.runs`** — Run-manifest indexes for the decomp-batch TUI (work item 0ff6bf0f): the
 - **`cjm_transcript_decomp_core.segments`** — Fine-segment inspection for the decomp TUI (work item 166dd2b8, half a):
 - **`cjm_transcript_decomp_core.state`** — Sidecar TUI state: last-used batch settings persisted across sessions (the
@@ -45,7 +48,9 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `assign_words_to_chunks` _function_ — Assign each FA word to a VAD chunk by timestamp overlap.
 - `build_segments_from_alignment` _function_ — Build a TextSegment per VAD chunk by grouping words by chunk assignment.
 - `carve_chunks_at_event_spans` _function_ — The event-carve stage (EVENT_SPLIT_POLICY, respine trial DEC 6cc10fb7):
+- `collapse_newlines` _function_ — The stored-text form of a slice that still holds a newline: a retained
 - `map_fa_words_to_text` _function_ — Map forced-alignment words back to character spans in the original text.
+- `normalize_external_text` _function_ — Offset-preserving fold-input normalisation for EXTERNAL variants (finding
 - `rescue_gap_words` _function_ — The word-rescue stage (WORD_RESCUE_POLICY, 96edc646 verdict bc7ece7b):
 - `sentence_end_word_indices` _function_ — Map capability-delivered sentence boundaries onto FA words (B.5: the
 - `split_chunks_at_sentence_gaps` _function_ — The sentence-split stage (SENTENCE_SPLIT_POLICY, DEC f1024568): refine the
@@ -57,7 +62,9 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `build_parser` _function_ — Build the CLI parser (subcommands: run).
 - `load_capabilities` _function_ — Discover manifests + load each requested capability (default instance).
 - `main` _function_ — CLI entry point (console script: `cjm-transcript-decomp-core`).
+- `respine_chunk_command` _function_ — Execute `respine-chunk` (work item 7a5e9c84, ruling 0b4d5cfa; the machinery lives
 - `run_command` _function_ — Execute the `run` subcommand: extend transcription-run manifest(s) with the fine spine.
+- `spine_command` _function_ — The spine retirement + compaction verbs (ruling a7617bd4; the machinery lives in
 
 ### `cjm_transcript_decomp_core.discovery`
 
@@ -111,6 +118,52 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `submit_and_wait` _function_ — Submit one capability job, wait for it, and return its result (raise on failure).
 - `vad_chunks_from_result` _function_ — Normalize a typed VAD result into segment-local VAD chunks.
 
+### `cjm_transcript_decomp_core.respine`
+
+- `bridge_edges` _function_ — Bridge NEXT from the previous live segment to the new first and from the new
+- `build_respine_op` _function_ — ONE replayed property-update op (pure): superseded_by on the old segments, the
+- `chunk_entry_for` _function_ — Resolve ONE coarse chunk of a source entry (pure): by manifest index, or by
+- `classify_dependents` _function_ — Sort the chunk's dependents into the two transferable classes and the rest
+- `decomp_config_from` _function_ — Rebuild the live spine's DecompConfig from its manifest snapshot (pure): the
+- `describe_dependents` _function_ — The refusal / readout listing (pure).
+- `find_decomp_manifest` _function_ — The decomp run that minted the live spine (pure): the newest manifest whose
+- `live_chunk_segments` _function_ — The chunk's LIVE segments: PART_OF its rendition, on the spine, not superseded.
+- `live_spine_index` _function_ — The whole live spine's (id, index) — the renumber plan's input.
+- `plan_renumber` _function_ — The renumber plan (pure; 0b4d5cfa (4), the 'do it properly' ruling): the new
+- `read_dependents` _function_ — What points at the old segments: every Correction with a CORRECTS edge into
+- `render_chunk_prompt` _function_ — Mode one (--prompt): the chunk's escalation prompt WITH CONTEXT — no writes.
+- `resolve_chunk_context` _function_ — Resolve the live spine -> its decomp manifest -> the transcription manifest it
+- `respine_chunk` _function_ — Mode two (--text-file): land -> re-derive the chunk under the live policy ->
+- `respine_op_id` _function_ — The op id `superseded_by` names: a function of (source, landed transcript).
+- `source_entry_for` _function_ — The transcription manifest's entry for a Source, by recomputed identity (pure).
+- `transfer_handler` _function_ — Discover the correction core's chunk-scoped transfer through the entry-point
+
+### `cjm_transcript_decomp_core.retire`
+
+- `annotate_spines` _function_ — Mark each spine row with its retirement state (pure; rows copied).
+- `apply_spine_fact` _function_ — Replay handler for spine-retire / spine-compaction / chunk-respine: property merges
+- `compact_retired` _function_ — The compact act over every retired, not-yet-compacted spine of the given sources:
+- `default_live_spine` _function_ — Which spine opens by default (pure). Preference is a DECLARED fact, never creation order:
+- `dependents_free` _function_
+- `dependents_map` _function_ — Every spine's dependents in one raw read (see DEPENDENTS_SQL); a spine absent from
+- `get_source` _function_ — The Source node as a dict (None when absent).
+- `journal_spine_retire` _function_ — Apply + journal one retirement fact (write-side dual of apply_spine_fact).
+- `list_sources` _function_ — Every Source (optionally one collection's members).
+- `list_spines` _function_ — The source's coexisting spines, grouped by skeleton hash and annotated with the retirement map.
+- `live_spines` _function_
+- `plan_retire` _function_ — Rules (a) + successor validation as a pure plan; the graph write is `journal_spine_retire`.
+- `plan_superseded` _function_ — The batch candidate list: every LIVE spine that is not the source's default (rule (c)) —
+- `resolve_source_id` _function_ — Resolve a human selector to ONE Source (refuses with the candidates).
+- `resolve_spine` _function_ — Resolve a picker-style selector to exactly one spine row (pure); refuses with the roster.
+- `retire_spine` _function_ — The whole retire act: list -> plan (rules a) -> dependents gate (rule b) -> journaled fact.
+- `retired_spines` _function_ — The Source's retirement map (pure): {spine key: {reason, successor, actor, ts[, compacted]}}.
+- `source_rendition_ids` _function_ — Every AudioRendition under a Source (all chains — retirement is per skeleton, not per chain).
+- `spine_dependents` _function_ — Rule (b)'s evidence: what on the graph points at the spine's segments.
+- `spine_fact_handlers` _function_
+- `spine_key` _function_
+- `spine_label` _function_
+- `spine_segment_ids` _function_
+
 ### `cjm_transcript_decomp_core.runs`
 
 - `DecompIndex` _class_ — Decomp-core run manifests read back: coverage chips for the batch stage
@@ -118,6 +171,7 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `SourceRunIndex` _class_ — Transcription-core run manifests — the decomp workflow's SOURCES — plus
 - `TrainingRunIndex` _class_ — Training-run manifests under the workspace training-runs/ dir — the
 - `group_batches` _function_ — Fold an ordered batch selection into headless hand-off groups.
+- `is_external_transcriber` _function_ — Whether a transcriber name is an external landing's (the `/manual` marker).
 
 ### `cjm_transcript_decomp_core.segments`
 
@@ -143,5 +197,5 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 
 ## Dependencies
 
-**Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `pyyaml`
-**Used by:** `cjm-transcript-decomp-qt`
+**Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `cjm-transcription-core`, `pyyaml`
+**Used by:** `cjm-transcript-correction-qt`, `cjm-transcript-decomp-qt`

{cjm_transcript_decomp_core-0.0.10 → cjm_transcript_decomp_core-0.0.20}/README.md

@@ -14,6 +14,8 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - **`cjm_transcript_decomp_core.launch`** — The shared launch surface every decomp shell drives through: the argument
 - **`cjm_transcript_decomp_core.models`** — Lean data shapes for the transcript-decomposition pipeline: in-core mirrors of the forced-alignment / VAD / text DTOs (no FastHTML deps), run configuration, the committed graph-segment carrier, and the decomposition run manifest (proto-bundle).
 - **`cjm_transcript_decomp_core.pipeline`** — The headless decomposition pipeline (stage 5: decomp is an EXTENDER). Load a transcription run manifest, verify the transcription-emitted graph root exists (the graph begins at transcription), then per source per pipeline-segment run VAD + per-transcriber forced alignment, build one aligned segment per VAD chunk with per-transcriber text variants, and attach the fine spine under the existing AudioSegment nodes via the layer's idempotent extend_graph — with HITL approval seams between alignment, commit, and the next source.
+- **`cjm_transcript_decomp_core.respine`** — Chunk RESPINE — re-derive ONE coarse chunk of a LIVE spine from a landed external transcript (work item 7a5e9c84; design ruling 0b4d5cfa, amendment 4a7ec4f8).
+- **`cjm_transcript_decomp_core.retire`** — Spine RETIREMENT + COMPACTION — safe removal of superseded decomposition spines (ruling a7617bd4, item eaefebd2).
 - **`cjm_transcript_decomp_core.runs`** — Run-manifest indexes for the decomp-batch TUI (work item 0ff6bf0f): the
 - **`cjm_transcript_decomp_core.segments`** — Fine-segment inspection for the decomp TUI (work item 166dd2b8, half a):
 - **`cjm_transcript_decomp_core.state`** — Sidecar TUI state: last-used batch settings persisted across sessions (the
@@ -25,7 +27,9 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `assign_words_to_chunks` _function_ — Assign each FA word to a VAD chunk by timestamp overlap.
 - `build_segments_from_alignment` _function_ — Build a TextSegment per VAD chunk by grouping words by chunk assignment.
 - `carve_chunks_at_event_spans` _function_ — The event-carve stage (EVENT_SPLIT_POLICY, respine trial DEC 6cc10fb7):
+- `collapse_newlines` _function_ — The stored-text form of a slice that still holds a newline: a retained
 - `map_fa_words_to_text` _function_ — Map forced-alignment words back to character spans in the original text.
+- `normalize_external_text` _function_ — Offset-preserving fold-input normalisation for EXTERNAL variants (finding
 - `rescue_gap_words` _function_ — The word-rescue stage (WORD_RESCUE_POLICY, 96edc646 verdict bc7ece7b):
 - `sentence_end_word_indices` _function_ — Map capability-delivered sentence boundaries onto FA words (B.5: the
 - `split_chunks_at_sentence_gaps` _function_ — The sentence-split stage (SENTENCE_SPLIT_POLICY, DEC f1024568): refine the
@@ -37,7 +41,9 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `build_parser` _function_ — Build the CLI parser (subcommands: run).
 - `load_capabilities` _function_ — Discover manifests + load each requested capability (default instance).
 - `main` _function_ — CLI entry point (console script: `cjm-transcript-decomp-core`).
+- `respine_chunk_command` _function_ — Execute `respine-chunk` (work item 7a5e9c84, ruling 0b4d5cfa; the machinery lives
 - `run_command` _function_ — Execute the `run` subcommand: extend transcription-run manifest(s) with the fine spine.
+- `spine_command` _function_ — The spine retirement + compaction verbs (ruling a7617bd4; the machinery lives in
 
 ### `cjm_transcript_decomp_core.discovery`
 
@@ -91,6 +97,52 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `submit_and_wait` _function_ — Submit one capability job, wait for it, and return its result (raise on failure).
 - `vad_chunks_from_result` _function_ — Normalize a typed VAD result into segment-local VAD chunks.
 
+### `cjm_transcript_decomp_core.respine`
+
+- `bridge_edges` _function_ — Bridge NEXT from the previous live segment to the new first and from the new
+- `build_respine_op` _function_ — ONE replayed property-update op (pure): superseded_by on the old segments, the
+- `chunk_entry_for` _function_ — Resolve ONE coarse chunk of a source entry (pure): by manifest index, or by
+- `classify_dependents` _function_ — Sort the chunk's dependents into the two transferable classes and the rest
+- `decomp_config_from` _function_ — Rebuild the live spine's DecompConfig from its manifest snapshot (pure): the
+- `describe_dependents` _function_ — The refusal / readout listing (pure).
+- `find_decomp_manifest` _function_ — The decomp run that minted the live spine (pure): the newest manifest whose
+- `live_chunk_segments` _function_ — The chunk's LIVE segments: PART_OF its rendition, on the spine, not superseded.
+- `live_spine_index` _function_ — The whole live spine's (id, index) — the renumber plan's input.
+- `plan_renumber` _function_ — The renumber plan (pure; 0b4d5cfa (4), the 'do it properly' ruling): the new
+- `read_dependents` _function_ — What points at the old segments: every Correction with a CORRECTS edge into
+- `render_chunk_prompt` _function_ — Mode one (--prompt): the chunk's escalation prompt WITH CONTEXT — no writes.
+- `resolve_chunk_context` _function_ — Resolve the live spine -> its decomp manifest -> the transcription manifest it
+- `respine_chunk` _function_ — Mode two (--text-file): land -> re-derive the chunk under the live policy ->
+- `respine_op_id` _function_ — The op id `superseded_by` names: a function of (source, landed transcript).
+- `source_entry_for` _function_ — The transcription manifest's entry for a Source, by recomputed identity (pure).
+- `transfer_handler` _function_ — Discover the correction core's chunk-scoped transfer through the entry-point
+
+### `cjm_transcript_decomp_core.retire`
+
+- `annotate_spines` _function_ — Mark each spine row with its retirement state (pure; rows copied).
+- `apply_spine_fact` _function_ — Replay handler for spine-retire / spine-compaction / chunk-respine: property merges
+- `compact_retired` _function_ — The compact act over every retired, not-yet-compacted spine of the given sources:
+- `default_live_spine` _function_ — Which spine opens by default (pure). Preference is a DECLARED fact, never creation order:
+- `dependents_free` _function_
+- `dependents_map` _function_ — Every spine's dependents in one raw read (see DEPENDENTS_SQL); a spine absent from
+- `get_source` _function_ — The Source node as a dict (None when absent).
+- `journal_spine_retire` _function_ — Apply + journal one retirement fact (write-side dual of apply_spine_fact).
+- `list_sources` _function_ — Every Source (optionally one collection's members).
+- `list_spines` _function_ — The source's coexisting spines, grouped by skeleton hash and annotated with the retirement map.
+- `live_spines` _function_
+- `plan_retire` _function_ — Rules (a) + successor validation as a pure plan; the graph write is `journal_spine_retire`.
+- `plan_superseded` _function_ — The batch candidate list: every LIVE spine that is not the source's default (rule (c)) —
+- `resolve_source_id` _function_ — Resolve a human selector to ONE Source (refuses with the candidates).
+- `resolve_spine` _function_ — Resolve a picker-style selector to exactly one spine row (pure); refuses with the roster.
+- `retire_spine` _function_ — The whole retire act: list -> plan (rules a) -> dependents gate (rule b) -> journaled fact.
+- `retired_spines` _function_ — The Source's retirement map (pure): {spine key: {reason, successor, actor, ts[, compacted]}}.
+- `source_rendition_ids` _function_ — Every AudioRendition under a Source (all chains — retirement is per skeleton, not per chain).
+- `spine_dependents` _function_ — Rule (b)'s evidence: what on the graph points at the spine's segments.
+- `spine_fact_handlers` _function_
+- `spine_key` _function_
+- `spine_label` _function_
+- `spine_segment_ids` _function_
+
 ### `cjm_transcript_decomp_core.runs`
 
 - `DecompIndex` _class_ — Decomp-core run manifests read back: coverage chips for the batch stage
@@ -98,6 +150,7 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 - `SourceRunIndex` _class_ — Transcription-core run manifests — the decomp workflow's SOURCES — plus
 - `TrainingRunIndex` _class_ — Training-run manifests under the workspace training-runs/ dir — the
 - `group_batches` _function_ — Fold an ordered batch selection into headless hand-off groups.
+- `is_external_transcriber` _function_ — Whether a transcriber name is an external landing's (the `/manual` marker).
 
 ### `cjm_transcript_decomp_core.segments`
 
@@ -123,5 +176,5 @@ A frontend-agnostic core for the transcript decomposition workflow — composes
 
 ## Dependencies
 
-**Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `pyyaml`
-**Used by:** `cjm-transcript-decomp-qt`
+**Depends on:** `cjm-capability-primitives`, `cjm-context-graph-layer`, `cjm-context-graph-primitives`, `cjm-substrate`, `cjm-transcript-graph-schema`, `cjm-transcription-core`, `pyyaml`
+**Used by:** `cjm-transcript-correction-qt`, `cjm-transcript-decomp-qt`

cjm_transcript_decomp_core-0.0.20/cjm_transcript_decomp_core/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.20"

{cjm_transcript_decomp_core-0.0.10 → cjm_transcript_decomp_core-0.0.20}/cjm_transcript_decomp_core/alignment.py

@@ -16,7 +16,50 @@ _PUNCT_RE = re.compile(r"[^\w\s]", re.UNICODE)
 # authoritative text — the v1-v3 token heuristic (closer/abbreviation/stub
 # lists) is RETIRED. The segmenter's identity (capability name + config hash)
 # joins the skeleton-identity composite beside this tag.
-SENTENCE_SPLIT_POLICY = "sentence-split/capability"
+# 'capability-v2' (finding efe88f17): the segmenter now reads EXTERNAL variants
+# through `normalize_external_text` — a pasted landing's wordwrap newlines
+# become spaces before the split, so pySBD (which splits at EVERY newline) no
+# longer cuts the skeleton mid-sentence; a paragraph break keeps ONE newline
+# and stays a boundary. Local transcriber text is untouched (it carries no
+# newlines). The split input changed for those variants, so the tag bumps.
+SENTENCE_SPLIT_POLICY = "sentence-split/capability-v2"
+
+# A newline run: its LAST newline is retained for a paragraph break; everything
+# else in the run becomes a space (length kept).
+_NEWLINE_RUN_RE = re.compile(r"[\r\n]+")
+_NEWLINE_INSIDE_RE = re.compile(r"[ \t]*[\r\n]+[ \t]*")
+
+
+def normalize_external_text(
+    text: str,  # An operator-landed external variant's text, verbatim as landed
+) -> str:  # The fold-input form: same length, wraps -> spaces, paragraph breaks -> one retained newline
+    """Offset-preserving fold-input normalisation for EXTERNAL variants (finding
+    efe88f17): the AI Studio 'Copy as text' gesture hard-wraps the transcript, and
+    pySBD splits at every newline — so a wrap mid-sentence became a sentence end
+    and a skeleton cut. Every newline run becomes spaces, except that a run
+    holding TWO OR MORE '\\n' (a paragraph / speaker break the model wrote on
+    purpose) keeps ONE '\\n' as its last character so the segmenter still sees
+    the boundary. '\\r' counts as whitespace of the run. The length never changes,
+    so every char-slice ref into the verbatim Transcript text stays valid; the
+    Transcript node itself is never rewritten — this is the fold's reading of it."""
+    def _run(m: "re.Match[str]") -> str:
+        run = m.group(0)
+        if run.count("\n") >= 2:
+            return " " * (len(run) - 1) + "\n"
+        return " " * len(run)
+    return _NEWLINE_RUN_RE.sub(_run, text)
+
+
+def collapse_newlines(
+    text: str,  # A segment's sliced text (may still hold a retained paragraph newline)
+) -> str:  # The same words, every newline run (with its flanking blanks) as one space
+    """The stored-text form of a slice that still holds a newline: a retained
+    paragraph break inside a segment (the split refused the cut, or no split ran)
+    is presentation, not content — it collapses to one space so the correction
+    app's <pre> walk lane never renders a ragged block. Slice refs are untouched
+    (they index the verbatim Transcript text, not this string)."""
+    return _NEWLINE_INSIDE_RE.sub(" ", text).strip()
+
 
 # Event-carve policy tag (respine trial DEC 6cc10fb7): versioned for the same
 # reason — it is a SKELETON IDENTITY input. 'v1' = cut-don't-label: model event
@@ -180,11 +223,14 @@ def build_segments_from_alignment(
     num_chunks: int,                # Total number of VAD chunks
     source_id: Optional[str] = None,           # Source row id for traceability
     source_provider_id: Optional[str] = None,  # Source provider identifier
+    collapse_newlines_in_text: bool = False,   # External variants (efe88f17): a newline still inside a slice -> one space in the stored text
 ) -> List[TextSegment]:  # One segment per VAD chunk
     """Build a TextSegment per VAD chunk by grouping words by chunk assignment.
 
     Each chunk's text is the original (punctuated) slice from the first to the
-    last word assigned to it; chunks with no words become empty segments.
+    last word assigned to it; chunks with no words become empty segments. With
+    `collapse_newlines_in_text` (external variants) the STORED text collapses any
+    newline run to one space — the slice refs stay the verbatim offsets.
     """
     chunk_spans: Dict[int, List[Tuple[int, int]]] = {}
     for span, chunk_idx in zip(spans, assignments):
@@ -197,6 +243,8 @@ def build_segments_from_alignment(
             seg_start = word_spans[0][0]
             seg_end = word_spans[-1][1]
             seg_text = text[seg_start:seg_end].strip()
+            if collapse_newlines_in_text:
+                seg_text = collapse_newlines(seg_text)
         else:
             seg_text = ""
             seg_start = None