Master

examples/aishell/s0/conf/train_conformer.yaml

@@ -41,7 +41,7 @@ dataset_conf:
         token_max_length: 200
         token_min_length: 1
     resample_conf:
-        resample_rate: 16000
+        resample_rate: 8000
     speed_perturb: true
     fbank_conf:
         num_mel_bins: 80

examples/aishell/s0/run.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/bin/bash /usr/bin/python
 
 # Copyright 2019 Mobvoi Inc. All Rights Reserved.
 . ./path.sh || exit 1;
@@ -47,6 +47,7 @@ train_set=train
 train_config=conf/train_conformer.yaml
 cmvn=true
 dir=exp/conformer
+#checkpoint=exp/conformer/9.pt
 checkpoint=
 num_workers=8
 prefetch=500
@@ -67,6 +68,7 @@ if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
   echo "stage -1: Data Download"
   local/download_and_untar.sh ${data} ${data_url} data_aishell
   local/download_and_untar.sh ${data} ${data_url} resource_aishell
+  readlink -f $(which python)
 fi
 
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
@@ -117,6 +119,9 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
   done
 fi
 
+export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+echo $CUDA_VISIBLE_DEVICES
+
 if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
   mkdir -p $dir
   # You have to rm `INIT_FILE` manually when you resume or restart a

runtime/core/bin/websocket_client_main.cc

@@ -32,7 +32,7 @@ int main(int argc, char* argv[]) {
   client.SendStartSignal();
 
   wenet::WavReader wav_reader(FLAGS_wav_path);
-  const int sample_rate = 16000;
+  const int sample_rate = 8000;
   // Only support 16K
   CHECK_EQ(wav_reader.sample_rate(), sample_rate);
   const int num_samples = wav_reader.num_samples();

runtime/web/streaming_app.py

@@ -7,7 +7,13 @@
 torch.manual_seed(777)  # for lint
 
 wenet.set_log_level(2)
-decoder = wenet.Decoder(lang='chs')
+decoder = wenet.Decoder(lang='chs',
+                        nbest=5,
+                        enable_timestamp=True,
+                        context=['宜享花','榕树','小辉贷'],
+                        context_score=3.0,
+                        model_dir='E:\\wenet\\pretrained\\runtime\\wenetspeech_u2pp_conformer_libtorch\\20220506_u2pp_conformer_libtorch')
+                        # model_dir='E:\\wenet\\pretrained\\runtime\\aishell_u2pp_conformer_libtorch\\aishell_u2pp_conformer_libtorch')
 
 def recognition(audio):
     sr, y = audio

tools/compute_cmvn_stats.py

@@ -1,4 +1,4 @@
-#!/usr/bin/env python3
+#!/usr/bin/python
 # encoding: utf-8
 
 import sys