From aa91adc60e1c4f3efae9b9e11789ed727df69460 Mon Sep 17 00:00:00 2001 From: Hui Zhang Date: Tue, 13 Dec 2022 20:08:40 +0800 Subject: [PATCH] update u2++ ctc prefix beam search result (#2736) --- speechx/build.sh | 2 +- .../examples/u2pp_ol/wenetspeech/RESULTS.md | 19 +++++++++++++++++++ 2 files changed, 20 insertions(+), 1 deletion(-) diff --git a/speechx/build.sh b/speechx/build.sh index e0a386752..7655f9635 100755 --- a/speechx/build.sh +++ b/speechx/build.sh @@ -20,4 +20,4 @@ fi mkdir -p build cmake -B build -DBOOST_ROOT:STRING=${boost_SOURCE_DIR} -cmake --build build +cmake --build build -j diff --git a/speechx/examples/u2pp_ol/wenetspeech/RESULTS.md b/speechx/examples/u2pp_ol/wenetspeech/RESULTS.md index 09584fd57..ef88357ee 100644 --- a/speechx/examples/u2pp_ol/wenetspeech/RESULTS.md +++ b/speechx/examples/u2pp_ol/wenetspeech/RESULTS.md @@ -6,8 +6,11 @@ > Intel(R) Xeon(R) Gold 6271C CPU @ 2.60GHz, support `avx512_vnni` > RTF with feature and decoder which is more end to end. + ### FP32 +`local/recognizer.sh` + #### CER ``` @@ -27,6 +30,8 @@ I1027 10:52:38.662876 51665 u2_recognizer_main.cc:123] RTF is: 0.309318 ### INT8 +`local/recognizer_quant.sh` + > RTF relative improve 12.8%, which count feature and decoder time. > Test under Paddle commit c331e2ce2031d68a553bc9469a07c30d718438f3 @@ -46,3 +51,17 @@ I1110 09:59:52.551712 37249 u2_recognizer_main.cc:122] total wav duration is: 36 I1110 09:59:52.551717 37249 u2_recognizer_main.cc:123] total decode cost:9737.63 sec I1110 09:59:52.551723 37249 u2_recognizer_main.cc:124] RTF is: 0.269674 ``` + +### CTC Prefix Beam Search + +`local/decode.sh` + +#### CER + +``` +Overall -> 6.74 % N=104765 C=98106 S=6516 D=143 I=401 +Mandarin -> 6.74 % N=104762 C=98106 S=6513 D=143 I=401 +English -> 0.00 % N=0 C=0 S=0 D=0 I=0 +Other -> 100.00 % N=3 C=0 S=3 D=0 I=0 + +```