docs(qwen3_tts): clarify voice cloning vs speaker synthesis usage rules

mlx_audio/tts/models/fish_qwen3_omni/fish_speech.py

@@ -407,6 +407,14 @@ def sample_rate(self) -> int:
     def model_type(self) -> str:
         return "fish_speech"
 
+    @classmethod
+    def model_quant_predicate(cls, path: str, module) -> bool:
+        import mlx.nn as nn
+        return (
+            not isinstance(module, nn.Embedding)
+            and "fast_" not in path
+        )
+
     def load_weights(self, weights, strict: bool = True):
         remapped = []
         for key, value in weights:
@@ -418,6 +426,9 @@ def load_weights(self, weights, strict: bool = True):
     def sanitize(self, weights: dict[str, mx.array]) -> dict[str, mx.array]:
         remapped = {}
         for key, value in weights.items():
+            if key.startswith("model."):
+                remapped[key] = value
+                continue
             if key.startswith("text_model.model."):
                 new_key = key[len("text_model.model.") :]
             elif key.startswith("audio_decoder."):

mlx_audio/tts/models/qwen3_tts/README.md

@@ -4,7 +4,9 @@ Alibaba's state-of-the-art multilingual TTS with three model variants.
 
 ## Voice Cloning
 
-Clone any voice using a reference audio sample. Provide the wav file and its transcript:
+Clone any voice using a reference audio sample. Provide the wav file and its exact transcript as a literal string (not a file path):
+
+> **Note:** Voice cloning is only supported on the **Base** models. Do not provide a `voice` (speaker name) when using `ref_audio` and `ref_text`, as it will cause a configuration conflict.
 
 ```python
 from mlx_audio.tts.utils import load_model