From c73d1a8e81582b91a9bdd6e82fce2e84f8d9d94b Mon Sep 17 00:00:00 2001
From: shixian.shi <shixian.shi@alibaba-inc.com>
Date: 星期一, 14 八月 2023 19:31:55 +0800
Subject: [PATCH] update func cif_wo_hidden

---
 funasr/bin/asr_inference_launch.py |    9 ++++++---
 1 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/funasr/bin/asr_inference_launch.py b/funasr/bin/asr_inference_launch.py
index fd47047..bc62b51 100644
--- a/funasr/bin/asr_inference_launch.py
+++ b/funasr/bin/asr_inference_launch.py
@@ -260,8 +260,6 @@
         hotword_list_or_file = None
         clas_scale = 1.0
 
-    if kwargs.get("device", None) == "cpu":
-        ngpu = 0
     if ngpu >= 1 and torch.cuda.is_available():
         device = "cuda"
     else:
@@ -1292,6 +1290,7 @@
         quantize_dtype: Optional[str] = "float16",
         streaming: Optional[bool] = False,
         simu_streaming: Optional[bool] = False,
+        full_utt: Optional[bool] = False,
         chunk_size: Optional[int] = 16,
         left_context: Optional[int] = 16,
         right_context: Optional[int] = 0,
@@ -1368,6 +1367,7 @@
         quantize_dtype=quantize_dtype,
         streaming=streaming,
         simu_streaming=simu_streaming,
+        full_utt=full_utt,
         chunk_size=chunk_size,
         left_context=left_context,
         right_context=right_context,
@@ -1418,7 +1418,7 @@
                         _end = (i + 1) * speech2text._ctx
 
                         speech2text.streaming_decode(
-                            speech[i * speech2text._ctx: _end], is_final=False
+                            speech[i * speech2text._ctx: _end + speech2text._right_ctx], is_final=False
                         )
 
                     final_hyps = speech2text.streaming_decode(
@@ -1426,6 +1426,8 @@
                     )
                 elif speech2text.simu_streaming:
                     final_hyps = speech2text.simu_streaming_decode(**batch)
+                elif speech2text.full_utt:
+                    final_hyps = speech2text.full_utt_decode(**batch)
                 else:
                     final_hyps = speech2text(**batch)
 
@@ -1814,6 +1816,7 @@
     group.add_argument("--ngram_weight", type=float, default=0.9, help="ngram weight")
     group.add_argument("--streaming", type=str2bool, default=False)
     group.add_argument("--simu_streaming", type=str2bool, default=False)
+    group.add_argument("--full_utt", type=str2bool, default=False)
     group.add_argument("--chunk_size", type=int, default=16)
     group.add_argument("--left_context", type=int, default=16)
     group.add_argument("--right_context", type=int, default=0)

--
Gitblit v1.9.1