add

examples/industrial_data_pretraining/sense_voice/demo.py

@@ -17,6 +17,7 @@ DecodingOptions = {
 	"task": ("ASR", "AED", "SER"),
 	"language": "auto",
 	"fp16": True,
+	"gain_event": True,
 	}
 
 res = model.generate(input=input_wav, batch_size_s=0, DecodingOptions=DecodingOptions)

funasr/models/paraformer/model.py

@@ -475,6 +475,8 @@ class Paraformer(torch.nn.Module):
         speech = speech.to(device=kwargs["device"])
         speech_lengths = speech_lengths.to(device=kwargs["device"])
         # Encoder
+        if kwargs.get("fp16", False):
+            speech = speech.half()
         encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
         if isinstance(encoder_out, tuple):
             encoder_out = encoder_out[0]

funasr/version.txt

@@ -1 +1 @@
-1.0.22
+1.0.23