Update onnx/builder.py

onnx/builder.py

@@ -127,7 +127,7 @@ def build_speech(args):
     dummy_inputs = (
         inputs["input_audio_embeds"],   # audio_embeds: torch.FloatTensor
         inputs["audio_attention_mask"], # audio_attention_mask: torch.BoolTensor
-        inputs["audio_embed_sizes"],    # audio_sizes: torch.FloatTensor
+        inputs["audio_embed_sizes"],    # audio_sizes: torch.LongTensor
         inputs["input_mode"],           # audio_projection_mode: int
     )
     dynamic_axes = {