From 30405d9acb534d2855f7f9e5d2fb93c5f9a317bc Mon Sep 17 00:00:00 2001 From: Hui Zhang Date: Fri, 15 Apr 2022 11:39:22 +0000 Subject: [PATCH] update aishell rsl --- speechx/examples/ds2_ol/aishell/.gitignore | 1 + speechx/examples/ds2_ol/aishell/README.md | 21 +++++++++++++++++++++ speechx/examples/ds2_ol/aishell/run.sh | 8 ++++---- 3 files changed, 26 insertions(+), 4 deletions(-) create mode 100644 speechx/examples/ds2_ol/aishell/README.md diff --git a/speechx/examples/ds2_ol/aishell/.gitignore b/speechx/examples/ds2_ol/aishell/.gitignore index bbd86a25..68f993b4 100644 --- a/speechx/examples/ds2_ol/aishell/.gitignore +++ b/speechx/examples/ds2_ol/aishell/.gitignore @@ -1,2 +1,3 @@ data exp +aishell_* diff --git a/speechx/examples/ds2_ol/aishell/README.md b/speechx/examples/ds2_ol/aishell/README.md new file mode 100644 index 00000000..eec67c3b --- /dev/null +++ b/speechx/examples/ds2_ol/aishell/README.md @@ -0,0 +1,21 @@ +# Aishell - Deepspeech2 Streaming + +## CTC Prefix Beam Search w/o LM + +``` +Overall -> 16.14 % N=104612 C=88190 S=16110 D=312 I=465 +Mandarin -> 16.14 % N=104612 C=88190 S=16110 D=312 I=465 +Other -> 0.00 % N=0 C=0 S=0 D=0 I=0 +``` + +## CTC Prefix Beam Search w LM + +``` + +``` + +## CTC WFST + +``` + +``` \ No newline at end of file diff --git a/speechx/examples/ds2_ol/aishell/run.sh b/speechx/examples/ds2_ol/aishell/run.sh index 7d65c8c2..3a1c19ee 100755 --- a/speechx/examples/ds2_ol/aishell/run.sh +++ b/speechx/examples/ds2_ol/aishell/run.sh @@ -65,7 +65,7 @@ cmvn-json2kaldi --json_file=$ckpt_dir/data/mean_std.json --cmvn_write_path=$cmvn ./local/split_data.sh $data $data/$aishell_wav_scp $aishell_wav_scp $nj -utils/run.pl JOB=1:$nj $data/split${nj}/JOB/feat_log \ +utils/run.pl JOB=1:$nj $data/split${nj}/JOB/feat.log \ linear-spectrogram-wo-db-norm-ol \ --wav_rspecifier=scp:$data/split${nj}/JOB/${aishell_wav_scp} \ --feature_wspecifier=ark,scp:$data/split${nj}/JOB/feat.ark,$data/split${nj}/JOB/feat.scp \ @@ -75,7 +75,7 @@ linear-spectrogram-wo-db-norm-ol \ text=$data/test/text # 4. recognizer -utils/run.pl JOB=1:$nj $data/split${nj}/JOB/log \ +utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.wolm.log \ ctc-prefix-beam-search-decoder-ol \ --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \ --model_path=$model_dir/avg_1.jit.pdmodel \ @@ -88,7 +88,7 @@ cat $data/split${nj}/*/result > ${label_file} utils/compute-wer.py --char=1 --v=1 ${label_file} $text > ${wer} # 4. decode with lm -utils/run.pl JOB=1:$nj $data/split${nj}/JOB/log_lm \ +utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.lm.log \ ctc-prefix-beam-search-decoder-ol \ --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \ --model_path=$model_dir/avg_1.jit.pdmodel \ @@ -111,7 +111,7 @@ fi # 5. test TLG decoder -utils/run.pl JOB=1:$nj $data/split${nj}/JOB/log_tlg \ +utils/run.pl JOB=1:$nj $data/split${nj}/JOB/recog.wfst.log \ wfst-decoder-ol \ --feature_rspecifier=scp:$data/split${nj}/JOB/feat.scp \ --model_path=$model_dir/avg_1.jit.pdmodel \