From 20241632ea905da3bfca98e52af2fdbfd98c21c6 Mon Sep 17 00:00:00 2001
From: Russell Ballestrini <russell@unturf.com>
Date: Sun, 9 Nov 2025 13:29:52 -0500
Subject: [PATCH] Fix Silero multilingual support with proper model loading
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Fixed Silero model caching to track language+speaker combination
- Updated Russian voices to use ru_v3 model (was v4_ru)
- Updated Spanish voices to use v3_es model (was v1_es)
- All model loading now properly switches between languages

Status:
✅ English (v3_en) - 119 voices working
✅ German (v3_de) - 6 voices working
✅ French (v3_fr) - 7 voices working
⚠️  Russian (ru_v3) - Model loading issue (investigating speaker format)
⚠️  Spanish (v3_es) - Model loading issue (investigating speaker format)

🦝 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
---
 speech.py                     | 10 +++++++---
 voice_to_speaker.default.yaml | 24 ++++++++++++------------
 2 files changed, 19 insertions(+), 15 deletions(-)

diff --git a/speech.py b/speech.py
index 9e26b74..9d0245e 100755
--- a/speech.py
+++ b/speech.py
@@ -477,10 +477,14 @@ async def generate_speech(request: GenerateSpeechRequest):
         speaker_id = voice_map.get('speaker', 'en_0')
         silero_speaker_key = voice_map.get('silero_speaker', 'v4_en')
 
-        # Load Silero model if not already loaded
-        if silero_model is None or silero_speakers.get(language) != silero_speaker_key:
+        # Create a unique key for this language+speaker combination
+        model_key = f"{language}_{silero_speaker_key}"
+
+        # Load Silero model if not already loaded or if language/speaker changed
+        if silero_model is None or silero_speakers.get('current') != model_key:
+            logger.info(f"Loading/switching Silero model to {language}/{silero_speaker_key}")
             silero_model = silero_wrapper(language=language, speaker=silero_speaker_key, device='cpu')
-            silero_speakers[language] = silero_speaker_key
+            silero_speakers['current'] = model_key
 
         # Generate audio
         audio_data = silero_model.tts(input_text, speaker_id=speaker_id)
diff --git a/voice_to_speaker.default.yaml b/voice_to_speaker.default.yaml
index cfbbc3d..089c95f 100644
--- a/voice_to_speaker.default.yaml
+++ b/voice_to_speaker.default.yaml
@@ -709,31 +709,31 @@ tts-1-silero:
     language: en
     speaker: random
     silero_speaker: v3_en
-  # Russian voices (v4_ru model)
+  # Russian voices (v3_ru/ru_v3 model) - use ru_v3 for better compatibility
   ru_aidar:
     language: ru
     speaker: aidar
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   ru_baya:
     language: ru
     speaker: baya
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   ru_kseniya:
     language: ru
     speaker: kseniya
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   ru_xenia:
     language: ru
     speaker: xenia
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   ru_eugene:
     language: ru
     speaker: eugene
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   ru_random:
     language: ru
     speaker: random
-    silero_speaker: v4_ru
+    silero_speaker: ru_v3
   # German voices (v3_de model)
   de_bernd_ungerer:
     language: de
@@ -759,23 +759,23 @@ tts-1-silero:
     language: de
     speaker: random
     silero_speaker: v3_de
-  # Spanish voices (v1_es model)
+  # Spanish voices (v3_es model)
   es_0:
     language: es
     speaker: es_0
-    silero_speaker: v1_es
+    silero_speaker: v3_es
   es_1:
     language: es
     speaker: es_1
-    silero_speaker: v1_es
+    silero_speaker: v3_es
   es_2:
     language: es
     speaker: es_2
-    silero_speaker: v1_es
+    silero_speaker: v3_es
   es_random:
     language: es
     speaker: random
-    silero_speaker: v1_es
+    silero_speaker: v3_es
   # French voices (v3_fr model)
   fr_0:
     language: fr