feat: add Qwen3-ASR speaker diarization; remove unnecessary punc_model from SPK examples

- Qwen3-ASR: support spk_model with forced_aligner for speaker diarization
- Qwen3-ASR: fix list input handling (numpy/tensor to tuple format)
- Qwen3-ASR: output timestamp as [start_ms, end_ms] for pipeline compatibility
- Qwen3-ASR: graceful fallback when forced_aligner not configured
- Remove punc_model from Fun-ASR-Nano/SenseVoice/Qwen3-ASR SPK demos and tests
  (these models output punctuation natively)
- Fix test scripts to handle both text/sentence keys in sentence_info
- auto_model: set return_time_stamps=True when spk_model present

examples/industrial_data_pretraining/fun_asr_nano/demo_spk.py

@@ -29,7 +29,6 @@ def main():
         vad_model="fsmn-vad",
         vad_kwargs={"max_single_segment_time": 30000},
         spk_model="cam++",
-        punc_model="ct-punc",
         device=device,
         hub="ms",
     )

examples/industrial_data_pretraining/sense_voice/demo_spk.py

@@ -28,7 +28,6 @@ def main():
         vad_model="fsmn-vad",
         vad_kwargs={"max_single_segment_time": 30000},
         spk_model="cam++",
-        punc_model="ct-punc",
         device=device,
     )
 

funasr/auto/auto_model.py

@@ -491,6 +491,7 @@ class AutoModel:
         self._reset_runtime_configs()
         if self.spk_model is not None and "output_timestamp" not in cfg:
             cfg["output_timestamp"] = True
+            cfg["return_time_stamps"] = True
         kwargs = self.kwargs
         # step.1: compute the vad model
         deep_update(self.vad_kwargs, cfg)

funasr/models/qwen3_asr/model.py

@@ -90,11 +90,18 @@ class Qwen3ASR(nn.Module):
         time1 = time.perf_counter()
 
         language = kwargs.get("language", None)
-        return_time_stamps = kwargs.get("return_time_stamps", False)
+        return_time_stamps = kwargs.get("return_time_stamps", False) or kwargs.get("output_timestamp", False)
         context = kwargs.get("context", "")
 
         if isinstance(data_in, (list, tuple)):
-            audio_inputs = list(data_in)
+            audio_inputs = []
+            for item in data_in:
+                if isinstance(item, np.ndarray):
+                    audio_inputs.append((item.astype(np.float32), 16000))
+                elif isinstance(item, torch.Tensor):
+                    audio_inputs.append((item.cpu().numpy().astype(np.float32), 16000))
+                else:
+                    audio_inputs.append(item)
         elif isinstance(data_in, str):
             audio_inputs = [data_in]
         elif isinstance(data_in, torch.Tensor):
@@ -114,6 +121,12 @@ class Qwen3ASR(nn.Module):
         time2 = time.perf_counter()
         meta_data["load_data"] = f"{time2 - time1:0.3f}"
 
+        # If timestamps requested but forced_aligner not configured, fall back to no timestamps
+        if return_time_stamps and self.qwen3_asr_model.forced_aligner is None:
+            logging.warning("return_time_stamps requires forced_aligner. Skipping timestamps. "
+                           "Initialize with forced_aligner='Qwen/Qwen3-ForcedAligner-0.6B' to enable.")
+            return_time_stamps = False
+
         results = self.qwen3_asr_model.transcribe(
             audio=audio_inputs,
             context=context,
@@ -132,7 +145,7 @@ class Qwen3ASR(nn.Module):
                 result_dict["language"] = r.language
             if return_time_stamps and r.time_stamps is not None:
                 result_dict["timestamp"] = [
-                    [ts.start_time, ts.end_time, ts.text]
+                    [int(ts.start_time), int(ts.end_time)]
                     for ts in r.time_stamps.items
                 ]
             output.append(result_dict)

tests_models/test_fun_asr_nano_spk.py

@@ -15,7 +15,6 @@ def main():
         vad_model="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
         vad_kwargs={"max_single_segment_time": 30000},
         spk_model="iic/speech_campplus_sv_zh-cn_16k-common",
-        punc_model="iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch",
         device="cpu",
         disable_update=True,
         hub="hf",
@@ -53,7 +52,7 @@ def main():
 
     print("[Fun-ASR-Nano-SPK] sentence_info:")
     for s in si:
-        print("  spk=%d | [%d-%d] %s" % (s["spk"], s["start"], s["end"], s["text"]))
+        print("  spk=%d | [%d-%d] %s" % (s["spk"], s["start"], s["end"], s.get("text", s.get("sentence", ""))))
 
     has_spk = all("spk" in s for s in si)
     if not has_spk:

tests_models/test_sensevoice_spk.py

@@ -14,7 +14,6 @@ def main():
         vad_model="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
         vad_kwargs={"max_single_segment_time": 30000},
         spk_model="iic/speech_campplus_sv_zh-cn_16k-common",
-        punc_model="iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch",
         device="cpu",
         disable_update=True,
     )
@@ -60,7 +59,7 @@ def main():
 
     print("[SenseVoice-SPK] sentence_info:")
     for s in si:
-        print("  spk=%d | [%d-%d] %s" % (s["spk"], s["start"], s["end"], rich_transcription_postprocess(s["text"])))
+        print("  spk=%d | [%d-%d] %s" % (s["spk"], s["start"], s["end"], rich_transcription_postprocess(s.get("text", s.get("sentence", "")))))
 
     has_spk = all("spk" in s for s in si)
     if not has_spk: