From 63e7a93ee4201f9ee55c06b001014e94afc5c685 Mon Sep 17 00:00:00 2001
From: Mikei386 <44135113+Mikei386@users.noreply.github.com>
Date: Sat, 5 Sep 2026 14:07:44 +0200
Subject: [PATCH] Improve German TTS range handling

---
 platform/docker/tts-gateway/test_tts_gateway.py |  8 ++++++++
 platform/docker/tts-gateway/tts_gateway.py      | 14 ++++++++++++++
 platform/hermes/025-cron-profile-root-fix       | 17 +++++++++++++++++
 3 files changed, 39 insertions(+)

diff --git a/platform/docker/tts-gateway/test_tts_gateway.py b/platform/docker/tts-gateway/test_tts_gateway.py
index 8ad4a53..1be3297 100644
--- a/platform/docker/tts-gateway/test_tts_gateway.py
+++ b/platform/docker/tts-gateway/test_tts_gateway.py
@@ -120,6 +120,14 @@ class LanguageSegmentationTests(unittest.TestCase):
         self.assertIn("2 Uhr 14", spoken)
         self.assertIn("1. September", spoken)
 
+    def test_qwen_speaks_temperature_and_time_ranges_with_bis(self):
+        spoken = gateway.prepare_for_qwen_speech(
+            "15°–23 °C, 93 %, ca. 12 mm, vor allem um 05–06 Uhr."
+        )
+        self.assertIn("15 bis 23 Grad", spoken)
+        self.assertIn("93 Prozent", spoken)
+        self.assertIn("5 bis 6 Uhr", spoken)
+
     def test_qwen_keeps_prosody_punctuation(self):
         spoken = gateway.prepare_for_qwen_speech(
             "Ist das gut? Ja! SarahTV: erreichbar."
diff --git a/platform/docker/tts-gateway/tts_gateway.py b/platform/docker/tts-gateway/tts_gateway.py
index b978034..8e6e98c 100644
--- a/platform/docker/tts-gateway/tts_gateway.py
+++ b/platform/docker/tts-gateway/tts_gateway.py
@@ -247,6 +247,20 @@ def normalize_for_german_speech(text: str) -> str:
         lambda match: f"{int(match.group(1))} Uhr {int(match.group(2))}",
         text,
     )
+    text = re.sub(
+        r"\b([01]?\d|2[0-3])\s*[-‐‑‒–—−]\s*"
+        r"([01]?\d|2[0-3])\s*Uhr\b",
+        lambda match: f"{int(match.group(1))} bis {int(match.group(2))} Uhr",
+        text,
+        flags=re.IGNORECASE,
+    )
+    text = re.sub(
+        r"(-?\d+(?:[,.]\d+)?)[ \t]*°?[ \t]*[-‐‑‒–—−][ \t]*"
+        r"(-?\d+(?:[,.]\d+)?)[ \t]*°[ \t]*(?:C)?",
+        r"\1 bis \2 Grad",
+        text,
+        flags=re.IGNORECASE,
+    )
     text = re.sub(
         r"(-?\d+(?:[,.]\d+)?)[ \t]*°[ \t]*(?:C)?[ \t]*/[ \t]*"
         r"(-?\d+(?:[,.]\d+)?)[ \t]*°[ \t]*(?:C)?",
diff --git a/platform/hermes/025-cron-profile-root-fix b/platform/hermes/025-cron-profile-root-fix
index 53fc24a..fa67682 100755
--- a/platform/hermes/025-cron-profile-root-fix
+++ b/platform/hermes/025-cron-profile-root-fix
@@ -15,3 +15,20 @@ cli_dir="${HERMES_HOME:-/opt/data}/.local/bin"
 mkdir -p "$cli_dir"
 ln -sfn /opt/hermes/.venv/bin/hermes "$cli_dir/hermes"
 ln -sfn /opt/hermes/.venv/bin/hermes /usr/local/bin/hermes
+
+# Hermes streams each detected sentence to TTS separately. Its generic
+# sentence splitter treats the period in the German abbreviation "ca." as a
+# sentence ending, causing "ca." and the following measurement to be spoken
+# as two unrelated utterances. Expand the abbreviation before splitting.
+tts_target=/opt/hermes/tools/tts_streaming.py
+tts_old='        self.buf = _THINK_BLOCK_RE.sub("", self.buf + delta)'
+tts_new='        self.buf = _THINK_BLOCK_RE.sub("", self.buf + delta)
+        self.buf = re.sub(r"\bca\.(?=\s)", "circa", self.buf, flags=re.IGNORECASE)'
+
+if [ -f "$tts_target" ] && grep -Fq "$tts_old" "$tts_target"; then
+    TTS_TARGET="$tts_target" TTS_OLD="$tts_old" TTS_NEW="$tts_new" python3 -c \
+        'import os; from pathlib import Path; p = Path(os.environ["TTS_TARGET"]); s = p.read_text(); p.write_text(s.replace(os.environ["TTS_OLD"], os.environ["TTS_NEW"], 1))'
+    echo "[tts-streaming-fix] expanded German ca. before sentence splitting"
+else
+    echo "[tts-streaming-fix] upstream code already fixed or layout changed; no action"
+fi