From 35b328c46c6a88290b4be59e75dc284a4c35261d Mon Sep 17 00:00:00 2001
From: Mubashir Osmani <mubashir@berri.ai>
Date: Sat, 18 Jul 2026 22:24:42 +0000
Subject: [PATCH] fix(dashscope): route qwen3-tts-vc tts to dashscope handler
 and cover speech dispatch

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/main.py                               |  4 ++-
 ...dashscope_text_to_speech_transformation.py | 27 +++++++++++++++++++
 2 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/litellm/main.py b/litellm/main.py
index e1ea631a4eb..db78d4453a0 100644
--- a/litellm/main.py
+++ b/litellm/main.py
@@ -7857,7 +7857,9 @@ def speech(
         Coroutine[Any, Any, HttpxBinaryResponseContent],
         None,
     ] = None
-    if custom_llm_provider == "openai" or custom_llm_provider in litellm.openai_compatible_providers:
+    if (
+        custom_llm_provider == "openai" or custom_llm_provider in litellm.openai_compatible_providers
+    ) and custom_llm_provider != "dashscope":
         if voice is None or not (isinstance(voice, str)):
             raise litellm.BadRequestError(
                 message="'voice' is required to be passed as a string for OpenAI TTS",
diff --git a/tests/test_litellm/llms/dashscope/test_dashscope_text_to_speech_transformation.py b/tests/test_litellm/llms/dashscope/test_dashscope_text_to_speech_transformation.py
index ad7c5b9c7c5..8c62b74a748 100644
--- a/tests/test_litellm/llms/dashscope/test_dashscope_text_to_speech_transformation.py
+++ b/tests/test_litellm/llms/dashscope/test_dashscope_text_to_speech_transformation.py
@@ -102,3 +102,30 @@ def test_transform_response_raises_without_audio_url(config: DashScopeTextToSpee
         config.transform_text_to_speech_response(
             model="qwen3-tts-vc", raw_response=raw, logging_obj=MagicMock()
         )
+
+
+def test_speech_dispatches_to_dashscope_handler():
+    import litellm
+
+    fake_audio = MagicMock()
+
+    with patch.object(
+        litellm.main.base_llm_http_handler,
+        "text_to_speech_handler",
+        return_value=fake_audio,
+    ) as mock_handler:
+        result = litellm.speech(
+            model="dashscope/qwen3-tts-vc",
+            input="hello world",
+            voice="my-cloned-voice",
+            api_key="sk-test",
+        )
+
+    assert result is fake_audio
+    mock_handler.assert_called_once()
+    kwargs = mock_handler.call_args.kwargs
+    assert kwargs["custom_llm_provider"] == "dashscope"
+    assert kwargs["model"] == "qwen3-tts-vc"
+    assert kwargs["voice"] == "my-cloned-voice"
+    assert isinstance(kwargs["text_to_speech_provider_config"], DashScopeTextToSpeechConfig)
+    assert kwargs["litellm_params"]["api_key"] == "sk-test"