Revert "add per template"

This reverts commit 9011b0e2faf24502f5bab8b920f57fea7a439357.

evaluation/svs_eval.py

@@ -72,7 +72,7 @@ def compute_dissonance_rate(intervals, dissonant_intervals={1, 2, 6, 10, 11}):
     return np.mean(dissonant) if intervals else np.nan
 
 
-def eval_per(audio_path, reference_text, model=None):
+def eval_per(audio_path, model=None):
     audio_array, sr = librosa.load(audio_path, sr=16000)
     # TODO: implement PER evaluation
     return {}
@@ -99,12 +99,12 @@ def load_evaluators(config):
     return loaded
 
 
-def run_evaluation(audio_path, evaluators, **kwargs):
+def run_evaluation(audio_path, evaluators):
     results = {}
     if "singmos" in evaluators:
         results.update(eval_singmos(audio_path, evaluators["singmos"]))
     if "per" in evaluators:
-        results.update(eval_per(audio_path, kwargs["llm_response"], evaluators["per"]))
+        results.update(eval_per(audio_path, evaluators["per"]))
     if "melody" in evaluators:
         results.update(eval_melody_metrics(audio_path, evaluators["melody"]))
     if "aesthetic" in evaluators:

interface.py

@@ -24,7 +24,6 @@ class GradioInterface:
             self.character_info[self.current_character].default_voice
         ]
         self.pipeline = SingingDialoguePipeline(self.default_config)
-        self.results = None
 
     def load_config(self, path: str):
         with open(path, "r") as f:
@@ -219,15 +218,14 @@ class GradioInterface:
             self.current_voice,
             output_audio_path=tmp_file,
         )
-        self.results = results
         formatted_logs = f"ASR: {results['asr_text']}\nLLM: {results['llm_text']}"
         return gr.update(value=formatted_logs), gr.update(
             value=results["output_audio_path"]
         )
 
     def update_metrics(self, audio_path):
-        if not audio_path or not self.results:
+        if not audio_path:
             return gr.update(value="")
-        results = self.pipeline.evaluate(audio_path, **self.results)
+        results = self.pipeline.evaluate(audio_path)
         formatted_metrics = "\n".join([f"{k}: {v}" for k, v in results.items()])
         return gr.update(value=formatted_metrics)