rtf

funasr/bin/vad_inference_launch.py

@@ -110,7 +110,8 @@ def inference_launch(mode, **kwargs):
     if mode == "offline":
         from funasr.bin.vad_inference import inference_modelscope
         return inference_modelscope(**kwargs)
-    elif mode == "online":
+    # elif mode == "online":
+    if "param_dict" in kwargs and kwargs["param_dict"]["online"]:
         from funasr.bin.vad_inference_online import inference_modelscope
         return inference_modelscope(**kwargs)
     else:

funasr/runtime/python/utils/test_rtf.py

@@ -0,0 +1,47 @@
+
+import time
+import sys
+import librosa
+backend=sys.argv[1]
+model_dir=sys.argv[2]
+wav_file=sys.argv[3]
+
+from torch_paraformer import Paraformer
+if backend == "onnxruntime":
+	from rapid_paraformer import Paraformer
+	
+model = Paraformer(model_dir, batch_size=1, device_id="-1")
+
+wav_file_f = open(wav_file, 'r')
+wav_files = wav_file_f.readlines()
+
+# warm-up
+total = 0.0
+num = 100
+wav_path = wav_files[0].split("\t")[1].strip() if "\t" in wav_files[0] else wav_files[0].split(" ")[1].strip()
+for i in range(num):
+	beg_time = time.time()
+	result = model(wav_path)
+	end_time = time.time()
+	duration = end_time-beg_time
+	total += duration
+	print(result)
+	print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
+
+# infer time
+beg_time = time.time()
+for i, wav_path_i in enumerate(wav_files):
+	wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+	result = model(wav_path)
+end_time = time.time()
+duration = (end_time-beg_time)*1000
+print("total_time_comput_ms: {}".format(int(duration)))
+
+duration_time = 0.0
+for i, wav_path_i in enumerate(wav_files):
+	wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+	waveform, _ = librosa.load(wav_path, sr=16000)
+	duration_time += len(waveform)/16.0
+print("total_time_wav_ms: {}".format(int(duration_time)))
+
+print("total_rtf: {:.5}".format(duration/duration_time))

funasr/runtime/python/utils/test_rtf.sh

@@ -0,0 +1,74 @@
+
+nj=64
+
+#:<<!
+backend=libtorch
+model_dir="/nfs/zhifu.gzf/export/damo/amp_int8/libtorch"
+tag=${backend}_fp32
+!
+
+:<<!
+backend=libtorch
+model_dir="/nfs/zhifu.gzf/export/damo/amp_int8/libtorch_fb20"
+tag=${backend}_amp_fb20
+!
+
+:<<!
+backend=onnxruntime
+model_dir="/nfs/zhifu.gzf/export/damo/amp_int8/onnx"
+tag=${backend}_fp32
+!
+
+:<<!
+backend=onnxruntime
+model_dir="/nfs/zhifu.gzf/export/damo/amp_int8/onnx_dynamic"
+tag=${backend}_fp32
+!
+
+scp=/nfs/haoneng.lhn/funasr_data/aishell-1/data/test/wav.scp
+scp="/nfs/zhifu.gzf/data_debug/test/wav_1500.scp"
+local_scp_dir=/nfs/zhifu.gzf/data_debug/test/${tag}/split$nj
+
+rtf_tool=test_rtf.py
+
+mkdir -p ${local_scp_dir}
+echo ${local_scp_dir}
+
+split_scps=""
+for JOB in $(seq ${nj}); do
+    split_scps="$split_scps $local_scp_dir/wav.$JOB.scp"
+done
+
+perl egs/aishell/transformer/utils/split_scp.pl $scp ${split_scps}
+
+
+for JOB in $(seq ${nj}); do
+  {
+    core_id=`expr $JOB - 1`
+    taskset -c ${core_id} python ${rtf_tool} ${backend} ${model_dir} ${local_scp_dir}/wav.$JOB.scp &> ${local_scp_dir}/log.$JOB.txt
+  }&
+
+done
+wait
+
+
+rm -rf ${local_scp_dir}/total_time_comput.txt
+rm -rf ${local_scp_dir}/total_time_wav.txt
+rm -rf ${local_scp_dir}/total_rtf.txt
+for JOB in $(seq ${nj}); do
+  {
+    cat ${local_scp_dir}/log.$JOB.txt | grep "total_time_comput" | awk -F ' '  '{print $2}' >> ${local_scp_dir}/total_time_comput.txt
+    cat ${local_scp_dir}/log.$JOB.txt | grep "total_time_wav" | awk -F ' '  '{print $2}' >> ${local_scp_dir}/total_time_wav.txt
+    cat ${local_scp_dir}/log.$JOB.txt | grep "total_rtf" | awk -F ' '  '{print $2}' >> ${local_scp_dir}/total_rtf.txt
+  }
+
+done
+
+total_time_comput=`cat ${local_scp_dir}/total_time_comput.txt | awk 'BEGIN {max = 0} {if ($1+0>max+0) max=$1 fi} END {print max}'`
+total_time_wav=`cat ${local_scp_dir}/total_time_wav.txt | awk '{sum +=$1};END {print sum}'`
+rtf=`awk 'BEGIN{printf "%.5f\n",'$total_time_comput'/'$total_time_wav'}'`
+speed=`awk 'BEGIN{printf "%.2f\n",1/'$rtf'}'`
+
+echo "total_time_comput_ms: $total_time_comput"
+echo "total_time_wav: $total_time_wav"
+echo "total_rtf: $rtf, speech: $speed"