admin can split a sentence into 2 sentences

sentence_api/tests/test_processing.py

@@ -1,6 +1,11 @@
 import numpy as np
 
-from sentence_api.processing import MOSS_ALGORITHM_VERSION, document_from_transcript
+from sentence_api.processing import (
+    MOSS_ALGORITHM_VERSION,
+    document_from_transcript,
+    split_sentence_text,
+)
+from sentence_api.models import SentenceBoundary
 from sentence_api.transcription import (
     Transcript,
     TranscriptionSegment,
@@ -13,6 +18,18 @@ SAMPLE_RATE = 16000
 VIDEO_HASH = "a" * 64
 
 
+class WordTimestampTranscriber:
+    def __init__(self, transcript):
+        self.transcript = transcript
+
+    @property
+    def available(self):
+        return True
+
+    def transcribe(self, audio_path, language=None):
+        return self.transcript
+
+
 def _speech(seconds: float) -> np.ndarray:
     t = np.arange(int(seconds * SAMPLE_RATE)) / SAMPLE_RATE
     return 0.25 * np.sin(2 * np.pi * 220 * t)
@@ -99,3 +116,45 @@ def test_document_from_transcript_splits_sentences_at_punctuation(tmp_path):
     assert document.sentences[1].start_ms == 9500
     assert document.sentences[1].end_ms == 24000
     assert all(sentence.reference_speech_duration_ms > 0 for sentence in document.sentences)
+
+
+def test_split_sentence_text_uses_whisper_word_timestamps(tmp_path):
+    transcript = Transcript(
+        text="Alpha beta. Gamma delta.",
+        segments=[
+            TranscriptionSegment(
+                start_seconds=0.0,
+                end_seconds=1.0,
+                text="Alpha beta. Gamma delta.",
+                words=[
+                    WordTimestamp(0.0, 0.25, "Alpha"),
+                    WordTimestamp(0.25, 0.5, "beta."),
+                    WordTimestamp(0.5, 0.75, "Gamma"),
+                    WordTimestamp(0.75, 1.0, "delta."),
+                ],
+            )
+        ],
+    )
+    audio_path = tmp_path / "sentence.wav"
+    audio_path.write_bytes(_wav_bytes(_speech(1.0), SAMPLE_RATE))
+    source = SentenceBoundary(
+        index=2,
+        start_ms=1000,
+        end_ms=5000,
+        text="Alpha beta Gamma delta",
+        language="en",
+        reference_speech_duration_ms=3800,
+    )
+
+    replacements = split_sentence_text(
+        transcriber=WordTimestampTranscriber(transcript),
+        media_path=audio_path,
+        sentence=source,
+        text="Alpha beta\nGamma delta",
+        language="en",
+    )
+
+    assert [(item.text, item.start_ms, item.end_ms) for item in replacements] == [
+        ("Alpha beta", 1000, 3000),
+        ("Gamma delta", 3000, 5000),
+    ]