fix: implement pronunciation adjustments for Palestinian past forms in TTS

app/main.py

@@ -34,7 +34,7 @@ from .reference import References
 from .scorer import Scorer, decode_audio, load_scorer
 from .speech_evaluation import SpeechEvaluationService
 from .texts import read_line, text_phrases
-from .tts import VOICES, SawtakTTS, TTSUnavailable, load_tts
+from .tts import VOICES, SawtakTTS, TTSUnavailable, load_tts, pronunciation_text
 from .verb_game import VerbExerciseGenerator
 
 logging.basicConfig(level=logging.INFO)
@@ -1127,7 +1127,7 @@ def create_app(scorer: Scorer | None = None, tts: SawtakTTS | None = None,
         names = {i: n for n, i in VOICES.items()}
         if voice and voice not in names:
             raise HTTPException(400, "unknown voice")
-        text = phrase_of(phrase_id)["variants"][0][0]
+        text = pronunciation_text(phrase_of(phrase_id)["variants"][0][0])
         if app.state.tts is None:
             raise HTTPException(503, "tts not configured")
         try:

app/tts.py

@@ -14,6 +14,7 @@ import io
 import json
 import logging
 import os
+import re
 import threading
 import urllib.error
 import urllib.parse
@@ -33,6 +34,28 @@ VOICES = {
     "חנאן": "e41781f5-127c-5eb0-b83c-50bf52a64caf",
 }
 
+# Unvowelled three-letter Palestinian past forms are often read by a TTS as MSA
+# (for example سمع as samaʿa).  Keep the course spelling untouched, but give the
+# voice the short vowels taught in the workbook's fiʿil pattern.
+_PAST_FIIL = {
+    "شرب": "شِرِب", "طلع": "طِلِع", "سمع": "سِمِع", "فهم": "فِهِم", "رجع": "رِجِع",
+    "نزل": "نِزِل", "وصل": "وِصِل", "ضحك": "ضِحِك", "لبس": "لِبِس", "لعب": "لِعِب",
+}
+_ARABIC_WORD = re.compile(r"[\u0621-\u064A\u066E-\u06D3]+")
+
+
+def pronunciation_text(text: str) -> str:
+    """Add only pronunciation-critical vowels for TTS; never changes displayed Arabic."""
+    def voiced(match: re.Match[str]) -> str:
+        word = match.group()
+        if word in _PAST_FIIL:
+            return _PAST_FIIL[word]
+        if word[:1] in ("و", "ف") and word[1:] in _PAST_FIIL:
+            return word[0] + _PAST_FIIL[word[1:]]
+        return word
+
+    return _ARABIC_WORD.sub(voiced, text)
+
 
 class TTSUnavailable(Exception):
     """No key configured, or the provider refused / failed. The message is safe to show."""

tests/test_scoring.py

@@ -213,6 +213,13 @@ def test_tts_endpoint_serves_phrase_audio_and_503_without_key():
         assert c.get("/api/health").json()["tts"] is True
 
 
+def test_tts_vowels_palestinian_fiil_past_without_changing_other_words():
+    from app.tts import pronunciation_text
+
+    assert pronunciation_text("هوّ سمع موسيقى، وبعدين رجع ولعب") == "هوّ سِمِع موسيقى، وبعدين رِجِع ولِعِب"
+    assert pronunciation_text("سمعت سمعو يسمع") == "سمعت سمعو يسمع"
+
+
 def test_spelling_variants_of_a_sound_are_not_errors():
     # the multilingual vocab spells one sound several ways; the model saying t̪ / ɾ / æ for t / r / a is not a mistake
     syms = SYMS + ["t", "t̪", "tː", "r", "ɾ", "æ", "b"]