From ffd4aac427ae2fece25bc2434da5f3f0fc8da0fc Mon Sep 17 00:00:00 2001
From: 仁迷 <haoneng.lhn@alibaba-inc.com>
Date: 星期五, 17 三月 2023 23:47:07 +0800
Subject: [PATCH] update paraformer streaming recipe
---
egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py | 8 ++++----
1 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py b/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py
index 3dfb912..c1c541b 100644
--- a/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py
@@ -13,7 +13,7 @@
model='damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online',
model_revision='v1.0.2')
-waveform, sample_rate = torchaudio.load("asr_example_zh.wav")
+waveform, sample_rate = torchaudio.load("waihu.wav")
speech_length = waveform.shape[1]
speech = waveform[0]
@@ -28,7 +28,7 @@
speech_cache = []
final_result = ""
-while len(speech_buffer) > 0:
+while len(speech_buffer) >= 960:
if first_chunk:
if len(speech_buffer) >= 14400:
rec_result = inference_pipeline(audio_in=speech_buffer[0:14400], param_dict=param_dict)
@@ -47,11 +47,11 @@
rec_result = inference_pipeline(audio_in=speech_buffer[:19200], param_dict=param_dict)
speech_buffer = speech_buffer[9600:]
else:
- cache_en["stride"] = len(speech_buffer) // 960
+ cache_en["stride"] = len(speech_buffer) // 960
cache_en["pad_right"] = 0
rec_result = inference_pipeline(audio_in=speech_buffer, param_dict=param_dict)
speech_buffer = []
- if rec_result['text'] != "sil":
+ if len(rec_result) !=0 and rec_result['text'] != "sil":
final_result += rec_result['text']
print(rec_result)
print(final_result)
--
Gitblit v1.9.1