An overrun occurred, which means the RTF of the current model on your board is larger than 1.

[MrJamesLZA]

芯片: rv1126
内核: 4.19
模型1: sherpa-ncnn-streaming-zipformer-20M-2023-02-17
模型2: sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23

出错-20M模型

[root@david:/home/asr]# /mnt/interference_ncnn_20m_en.sh  
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt", encoder num_threads=1, decoder num_threads=1, joiner num_threads=1), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.2, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=300)), enable_endpoint=True, hotwords_file="greedy_search", hotwrods_score=1.5)
Current sample rate: 16000
Recording started!
Use recording device: plughw:0,0
An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 

脚本内容

[root@david:/home/asr]# cat /mnt/interference_ncnn_20m_en.sh 
#!/bin/sh 

cd /home/asr

sherpa-ncnn-alsa \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin \
  plughw:0,0 \
  1 \
  greedy_search

[root@david:/home/asr]# ls
interference_ncnn_14m.sh
sherpa-ncnn-streaming-zipformer-20M-2023-02-17
sherpa-ncnn-streaming-zipformer-bilingual-zh-en-2023-02-13
sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16
sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23

cpu信息

[root@david:/home/asr]# uname -a
Linux localhost 4.19.111 #148 SMP PREEMPT Mon Mar 24 19:19:35 CST 2025 armv7l GNU/Linux
[root@david:/home/asr]# 
[root@david:/home/asr]# cat /proc/cpuinfo 
processor	: 0
model name	: ARMv7 Processor rev 5 (v7l)
BogoMIPS	: 83.76
Features	: half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae 
CPU implementer	: 0x41
CPU architecture: 7
CPU variant	: 0x0
CPU part	: 0xc07
CPU revision	: 5

processor	: 1
model name	: ARMv7 Processor rev 5 (v7l)
BogoMIPS	: 83.76
Features	: half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae 
CPU implementer	: 0x41
CPU architecture: 7
CPU variant	: 0x0
CPU part	: 0xc07
CPU revision	: 5

processor	: 2
model name	: ARMv7 Processor rev 5 (v7l)
BogoMIPS	: 83.76
Features	: half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae 
CPU implementer	: 0x41
CPU architecture: 7
CPU variant	: 0x0
CPU part	: 0xc07
CPU revision	: 5

processor	: 3
model name	: ARMv7 Processor rev 5 (v7l)
BogoMIPS	: 83.76
Features	: half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae 
CPU implementer	: 0x41
CPU architecture: 7
CPU variant	: 0x0
CPU part	: 0xc07
CPU revision	: 5

Hardware	: Generic DT based system
Revision	: 0000
Serial		: e9b87daeab52b98c
[root@david:/home/asr]# 

14M模型单转wav示例文件

[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# for method in greedy_search modified_beam_search; do
>   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
>     2 \
>     $method
> done
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=2, decoder num_threads=2, joiner num_threads=2), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.76 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.64 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 3.309 s
Real time factor (RTF): 3.309 / 5.611 = 0.590
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=2, decoder num_threads=2, joiner num_threads=2), decoder_config=DecoderConfig(method="modified_beam_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.72 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.56 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 5.258 s
Real time factor (RTF): 5.258 / 5.611 = 0.937
[root@david:/home/asr]# 

参考视频

https://www.bilibili.com/video/BV1Lo4y1e77y/?spm_id_from=333.337.search-card.all.click&vd_source=b0da736c0565522ea22f0adf3b77c6e9
https://www.bilibili.com/video/BV1K84y1K76C/?spm_id_from=333.337.search-card.all.click&vd_source=b0da736c0565522ea22f0adf3b77c6e9

---

综合:

总是还没说话就直接提示overrun...是什么原因呢?
如果说sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16和sherpa-ncnn-streaming-zipformer-bilingual-zh-en-2023-02-13单转wav的RTF都>1, 导致了overrun, 但是14M和20M的单转RTF都已经< 1了.

不知道要如何进一步优化, any response is appreciated

[csukuangfj]

[MrJamesLZA]

[root@david:/home/asr]# ls
sherpa-ncnn-streaming-zipformer-20M-2023-02-17
sherpa-ncnn-streaming-zipformer-bilingual-zh-en-2023-02-13
sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16
sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23
[root@david:/home/asr]# /mnt/interference_ncnn_14m.sh 
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.2, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=300)), enable_endpoint=True, hotwords_file="greedy_search", hotwrods_score=1.5)
Current sample rate: 16000
Recording started!
Use recording device: plughw:1,0
0:是啊嗯
1:我我
2:是
3:啊嗯An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# /mnt/interference_ncnn_14m.sh 
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.2, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=300)), enable_endpoint=True, hotwords_file="greedy_search", hotwrods_score=1.5)
Current sample rate: 16000
Recording started!
Use recording device: plughw:1,0
0:呃An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# /mnt/interference_ncnn_14m.sh 
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.2, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=300)), enable_endpoint=True, hotwords_file="greedy_search", hotwrods_score=1.5)
Current sample rate: 16000
Recording started!
Use recording device: plughw:1,0
0:啊是啊嗯
1:我我An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# /mnt/interference_ncnn_14m.sh 
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.2, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=300)), enable_endpoint=True, hotwords_file="greedy_search", hotwrods_score=1.5)
Current sample rate: 16000
Recording started!
Use recording device: plughw:0,0
0:我是An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# 
[root@david:/home/asr]# cat /mnt/interference_ncnn_14m.sh 
#!/bin/sh

cd /home/asr

sherpa-ncnn-alsa \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
  plughw:0,0 \
  4 \
  greedy_search

[root@david:/home/asr]# 

我改了...识别都是不对的, 然后好像是有点阻塞, 然后卡住过几秒钟就overrun了...跟麦克风会有关系吗?我录制过arecord -D hw:0,0 -f S16_LE -r 48000 -c 2 -d 5 test.wav,声音是ok的.

[MrJamesLZA]

用num threads=4的RTF数值更低 0.498

[root@david:/home/asr]# for method in greedy_search modified_beam_search; do
>   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
>     4 \
>     $method
> done
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.76 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.64 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 2.792 s
Real time factor (RTF): 2.792 / 5.611 = 0.498
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="modified_beam_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.72 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.56 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 4.675 s
Real time factor (RTF): 4.675 / 5.611 = 0.833
[root@david:/home/asr]# 

[MrJamesLZA]

我写了一个bin来做debug

#include <csignal>
#include <cstdio>  // 使用 cstdio 以便和 C-API 风格保持一致
#include <string>
#include <vector>

// Sherpa-NCNN 的 C-API 头文件
#include "sherpa-ncnn/c-api/c-api.h"

// 你项目中的 Alsa 类的头文件 (这是一个 C++ 类)
#include "alsa.h"

// --- 全局变量用于信号处理 ---
bool g_stop = false;

void SignalHandler(int sig) {
	g_stop = true;
	fprintf(stderr, "\nCaught Ctrl + C. Exiting...\n");
}

int main() {
	// 注册 Ctrl+C 信号处理器
	signal(SIGINT, SignalHandler);

	// 1. 配置 Recognizer (使用 C API 结构体)
	SherpaNcnnRecognizerConfig recognizer_config;
	memset(&recognizer_config, 0, sizeof(recognizer_config));

	// 特征配置
	recognizer_config.feat_config.sampling_rate = 16000.0f;
	recognizer_config.feat_config.feature_dim = 80;

	// // 模型配置 (使用相对路径, 假设模型文件夹和可执行文件在同一级目录)
	// const char* model_dir = "./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/";
	// char encoder_param[256], encoder_bin[256], decoder_param[256], decoder_bin[256],
	//     joiner_param[256], joiner_bin[256], tokens[256];

	// snprintf(encoder_param, sizeof(encoder_param), "%s/encoder_jit_trace-pnnx.ncnn.param",
	//          model_dir);
	// snprintf(encoder_bin, sizeof(encoder_bin), "%s/encoder_jit_trace-pnnx.ncnn.bin", model_dir);
	// snprintf(decoder_param, sizeof(decoder_param), "%s/decoder_jit_trace-pnnx.ncnn.param",
	//          model_dir);
	// snprintf(decoder_bin, sizeof(decoder_bin), "%s/decoder_jit_trace-pnnx.ncnn.bin", model_dir);
	// snprintf(joiner_param, sizeof(joiner_param), "%s/joiner_jit_trace-pnnx.ncnn.param", model_dir);
	// snprintf(joiner_bin, sizeof(joiner_bin), "%s/joiner_jit_trace-pnnx.ncnn.bin", model_dir);
	// snprintf(tokens, sizeof(tokens), "%s/tokens.txt", model_dir);

	// 模型配置 (使用相对路径, 假设模型文件夹和可执行文件在同一级目录)
	// 修改点 1: 更新模型目录名称
	const char* model_dir = "/home/asr/sherpa-ncnn-conv-emformer-transducer-small-2023-01-09";

	// 用于存储完整文件路径的字符数组
	char encoder_param[256], encoder_bin[256], decoder_param[256], decoder_bin[256],
	    joiner_param[256], joiner_bin[256], tokens[256];

	// --- 根据新的模型文件和 int8 优先规则，生成文件路径 ---

	// 修改点 2: Encoder 使用 int8 模型
	snprintf(encoder_param, sizeof(encoder_param), "%s/encoder_jit_trace-pnnx.ncnn.int8.param",
	         model_dir);
	snprintf(encoder_bin, sizeof(encoder_bin), "%s/encoder_jit_trace-pnnx.ncnn.int8.bin",
	         model_dir);

	// 修改点 3: Decoder 没有 int8 版本, 使用普通模型
	snprintf(decoder_param, sizeof(decoder_param), "%s/decoder_jit_trace-pnnx.ncnn.param",
	         model_dir);
	snprintf(decoder_bin, sizeof(decoder_bin), "%s/decoder_jit_trace-pnnx.ncnn.bin", model_dir);

	// 修改点 4: Joiner 使用 int8 模型
	snprintf(joiner_param, sizeof(joiner_param), "%s/joiner_jit_trace-pnnx.ncnn.int8.param",
	         model_dir);
	snprintf(joiner_bin, sizeof(joiner_bin), "%s/joiner_jit_trace-pnnx.ncnn.int8.bin", model_dir);

	// tokens 文件路径保持不变
	snprintf(tokens, sizeof(tokens), "%s/tokens.txt", model_dir);

	recognizer_config.model_config.encoder_param = encoder_param;
	recognizer_config.model_config.encoder_bin = encoder_bin;
	recognizer_config.model_config.decoder_param = decoder_param;
	recognizer_config.model_config.decoder_bin = decoder_bin;
	recognizer_config.model_config.joiner_param = joiner_param;
	recognizer_config.model_config.joiner_bin = joiner_bin;
	recognizer_config.model_config.tokens = tokens;

	recognizer_config.model_config.num_threads = 2;  // 根据你的板子性能调整
	recognizer_config.model_config.use_vulkan_compute = 0;

	// 解码器配置
	recognizer_config.decoder_config.decoding_method = "greedy_search";

	// 端点检测配置
	recognizer_config.enable_endpoint = 1;  // 启用端点检测
	recognizer_config.rule1_min_trailing_silence = 2.4f;
	recognizer_config.rule2_min_trailing_silence = 1.2f;
	recognizer_config.rule3_min_utterance_length = 300;

	// 2. 初始化 Recognizer (使用 C-API 函数)
	SherpaNcnnRecognizer* recognizer = CreateRecognizer(&recognizer_config);
	if (!recognizer) {
		fprintf(stderr, "Failed to create recognizer. Check your model paths.\n");
		return -1;
	}

	// 3. 初始化 ALSA 音频输入 (使用你提供的 C++ 类)
	const char* device_name = "plughw:0,0";  // 确保设备名正确
	Alsa alsa(device_name);

	// 4. 创建识别流 (使用 C-API 函数)
	SherpaNcnnStream* stream = CreateStream(recognizer);

	// 5. 开始主循环
	fprintf(stderr, "\nListening... Press Ctrl+C to exit.\n\n");
	std::string last_text;
	int32_t segment_index = 0;

	// 每次从声卡读取 0.1 秒的数据
	int32_t samples_per_read = static_cast<int32_t>(alsa.GetActualSampleRate() * 0.1);

	while (!g_stop) {
		// 从 Alsa 对象读取音频。
		// 注意：返回的 samples 已经是被 alsa 对象内部重采样到 16000Hz 的了
		const std::vector<float>& samples = alsa.Read(samples_per_read);

		fprintf(stdout, "\n-------------\npos.1\n");
		fflush(stdout);

		// 将数据送入识别器
		AcceptWaveform(stream, alsa.GetExpectedSampleRate(), samples.data(), samples.size());

		fprintf(stdout, "pos.2\n");
		fflush(stdout);
		// 准备好就解码
		while (IsReady(recognizer, stream)) {
			Decode(recognizer, stream);
			fprintf(stdout, "pos.3\n");
			fflush(stdout);
		}
		fprintf(stdout, "pos.4\n");
		fflush(stdout);

		// 获取结果
		const SherpaNcnnResult* result = GetResult(recognizer, stream);
		fprintf(stdout, "pos.5\n");
		fflush(stdout);
		if (result->text && strlen(result->text) > 0) {
			if (last_text != result->text) {
				last_text = result->text;
				// 用 \r 和空格刷新当前行，实现动态显示效果
				fprintf(stderr, "\rSegment %d: %s", segment_index, last_text.c_str());
			}
		}
		fprintf(stdout, "pos.6\n");
		fflush(stdout);
		// 必须释放 C-API 返回的 result 对象，否则内存泄漏
		DestroyResult(result);

		fprintf(stdout, "pos.7\n");
		fflush(stdout);

		// 检查是否检测到语音端点
		if (IsEndpoint(recognizer, stream)) {
			if (!last_text.empty()) {
				fprintf(stderr, "\n");  // 换行，确认上一句话
				segment_index++;
			}
			// 重置流以准备下一次识别
			Reset(recognizer, stream);
			last_text.clear();
		}
		fprintf(stdout, "pos.8\n");
		fflush(stdout);
	}

	fprintf(stderr, "\nCleaning up...\n");

	// 6. 释放资源
	DestroyStream(stream);
	DestroyRecognizer(recognizer);

	return 0;
}

程序运行的log如下

[root@david:/home/asr]# ./04.07-rv1126-alsa/bin/prototype.alsa 
Current sample rate: 16000
Recording started!

Listening... Press Ctrl+C to exit.


-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.3
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.3
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.3
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.3
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.4
pos.5
pos.6
pos.7
pos.8

-------------
pos.1
pos.2
pos.3
pos.4
pos.5
pos.6
pos.7
pos.8
An overrun occurred, which means the RTF of the current model on your board is larger than 1. You can use ./bin/sherpa-ncnn to verify that. Please select a smaller model whose RTF is less than 1 for your board.[root@david:/home/asr]# 

不知这样的流程是否有什么异常, 包要isReady才能取, 但是很快就出现overrun,是缓冲区溢出了吗?还是?

CLI下和自写bin都试过,从1~4的num threads都试过, 暂时都是几秒钟就出现overrun...

[MrJamesLZA]

还有,不知道rv1126有没有rknn的成功先例...谢谢

[MrJamesLZA]

以下是第二个奇怪的部分: RTF都高得离谱. 不知道如何能够复现 https://www.bilibili.com/video/BV1K84y1K76C/?spm_id_from=333.337.search-card.all.click&vd_source=b0da736c0565522ea22f0adf3b77c6e9视频里的这种速度和效率?我们的CPU都是rv1126, 是有什么其他部分会导致RTF这么高吗?这么高的RTF是否也无法比较好的运行sherpa-ncnn-alsa?

以下是我这两天做的几个验证:

conv-emformer-transducer-small-int8量化模型

命令

sherpa-ncnn \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/test_wavs/1089-134686-0001.wav 

执行结果(rtf=1.2)

[root@david:/home/asr]# sherpa-ncnn \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/test_wavs/1089-134686-0001.wav \
> 
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param", encoder_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin", decoder_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param", joiner_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin", tokens="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/test_wavs/1089-134686-0001.wav
wav duration (s): 6.625
Started!
Done!
Recognition result for ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/test_wavs/1089-134686-0001.wav
text:  AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BRAFFLES
timestamps: 0.32 0.96 1.04 1.12 1.4 1.6 1.64 1.8 1.96 2.08 2.2 2.24 2.44 2.56 2.6 2.68 2.8 3 3.28 3.52 3.6 3.76 3.96 4.36 4.52 4.56 4.76 4.84 5.08 5.12 5.2 5.24 5.48 5.64 5.84 5.88 5.96 6.12 6.24 
Elapsed seconds: 8.481 s
Real time factor (RTF): 8.481 / 6.625 = 1.280
[root@david:/home/asr]# 

推理bbc音频-命令

sherpa-ncnn \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param \
  ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin \
  /home/asr/media/english-bbc-news/english-bbc-news.wav

执行结果(rtf=1.3)

[root@david:/home/asr]# sherpa-ncnn \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param \
>   ./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin \
>   /home/asr/media/english-bbc-news/english-bbc-news.wav
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.param", encoder_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/encoder_jit_trace-pnnx.ncnn.int8.bin", decoder_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.param", joiner_bin="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/joiner_jit_trace-pnnx.ncnn.int8.bin", tokens="./sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)

wav filename: /home/asr/media/english-bbc-news/english-bbc-news.wav
wav duration (s): 142.64
Started!
Done!
Recognition result for /home/asr/media/english-bbc-news/english-bbc-news.wav
text:  BUT NOW INDIAN ADMINISTERED KUSHMA WHERE POLICE HAVE NAMED THREE MILITANT SUSPECTED OF CARRYING OUT TUESDAY'S DEADLY ATTACKING THE REGION THEY SAY THAT TWO ARE PAGUSTINEE NATIONALS THE POLICE OF ALSO ANNOUNCED A REWARD OF MORE THAN TWENTY THOUSAND DOLLARS FOR INFORMATION ON THE ATTACK IN WHICH TWENTY SIX PEOPLE WERE KILLED AROUND FIFTEEN HUNDRED PEOPLE IN THE REGION HAVE BEEN DETAINED WHERE WE CAN SPEAK TO OUR SOUTH ASIA CORRESPONDENT YAGITTA LE MAI WHO JOINS US FROM SWINEGAR IN INDIAN ADMINISTERED CASHMERE SO JUST AND BRING US UP TO DATE THREE PETENTIAL SUSPECTS NAMED AH THAT'S RIGHT WHAT OPERALY SOURCES HAVE TOLD US THAT THEY BELIEVE THERE WERE A TOTAL OF FOUR MILITANTS INVOLGANY ATTACK WHICH INCLUDE A PEOPLE THEY SAY WHO ARE LOCALS BUT ALSO PEOPLE WHO ARE FOREIGN NATIONALS ARE POLICES OF ALTERED ALL AS THAT FIFTEEN HUNDRED PEOPLE FROM ACROSS THE CASHMI REGION HAVE BEEN ROUNDED UP IN CONNECTION WITH THE ATTACKS THEY SAY THESE ARE PEOPLE VASE SUSPECT WHO REMAIN IN CONNECTION WITH TERRA NETWARKS AND THEY BELIEVE THEY MIGHT GIVE THEM CLUES AS TO WHO CARRIED OUT THIS OPERATION AND HOW IT WAS CARRIED OUT THE SEARCH FOR THE ATTACKERS OF COURSE CONTINUES THE AREA OF PAHG CONTINUES TO BE SHUT DOWN AN ORDANT OFF ARE THEIR RESERTUBILIAN RUPEE REWARD OR TWENTY THREE A THOUSAND DOLLARS BEING OFFERED FOR ANY ONE WHO CAN COME UP WITH ANY INFORMATION ABOUT THOSE ATTACKERS AS FAR AS REST CONCERNED REMAINS TENSE BUT AFTER THE SHOT DOWN YESTERDAY LIFE IS SLOWLY RESUMING TO NORMAL SHOP'S BUSINESSES SCHOOLS ARE IN THIS AREA HAVE REOPENED IN THE YESTERDAY RENOUNCE ER IN ESSAY WE SAW ANNOUNCED A WRATH OF MEASURES AGAINST BARKISH THORN A WITCH APPEARED ARE TO BE DIRECTED AT HITTING IT ECONOMICALLY AT THE KEY MEASURES OF COURSE IS THE CLOSING DOWN OF THE MAIN BORDER CROSSING BETWEEN THE TWO COUNTRIES WHICH WILL AFFECT TRADE BETWEEN INDIA AND PARKASON THE OTHER SIGNIFICANT ANNOUNCEMENT WAS THE SUSPENSION OF A RIVER WATER'S SHARINGT TREATY WHICH WAS SIGNED BACK IN NINETEEN SIXTY WHICH WILL HAVE AN IMPACT ON THE WATER AVAILABLE FOR IRRIGATION THE AGRICULTURAL LAND APART FROM THAT INDIA ALSO ANNOUNCED THE EXPULSION OF DIPLOMAT AND ALTER SAID IT WAS WITHDRAWING A SUM OF ITS OWN DIPLOMATS FROM ISLAMABOD
timestamps: 0.32 0.68 1.96 2.08 2.12 2.32 2.4 2.44 2.52 2.6 2.76 2.84 2.96 3 3.08 3.16 3.24 3.52 3.72 3.88 3.96 4.16 4.32 4.48 4.76 5.16 5.28 5.44 5.56 5.76 5.88 5.96 6.04 6.12 6.32 6.48 6.68 6.72 6.84 7.08 7.12 7.2 7.32 7.36 7.44 7.52 7.56 7.6 7.72 7.84 8.04 8.16 8.24 8.32 8.48 8.64 8.76 8.8 9.28 9.48 9.76 10 10.32 10.48 10.6 10.68 10.84 10.96 11.08 11.2 11.32 11.44 11.48 12.04 12.16 12.28 12.36 12.48 12.6 12.88 13.04 13.08 13.12 13.16 13.2 13.32 13.52 13.6 13.84 14 14.12 14.32 14.36 14.48 14.6 14.8 14.84 15 15.04 15.24 15.32 15.4 15.52 15.68 15.84 16 16.04 16.12 16.16 16.4 16.56 16.72 16.84 16.96 17.28 17.44 17.64 17.68 17.72 17.84 17.96 18.24 18.6 18.84 18.88 19 19.08 19.52 19.68 19.72 19.84 20 20.08 20.12 20.32 20.68 20.92 21.04 21.16 21.32 21.36 21.72 21.96 22.48 22.64 22.8 22.88 23.36 23.48 23.6 23.76 24 24.2 24.4 24.52 24.56 24.68 24.88 24.92 25 25.04 25.12 25.16 25.24 25.28 25.4 25.44 25.72 25.76 25.8 25.84 25.96 26.12 26.32 26.52 26.72 26.88 27.12 27.28 27.4 27.56 27.8 28 28.12 28.24 28.28 28.36 29.08 29.28 29.4 29.44 29.72 29.8 29.84 29.92 29.96 30.16 30.24 30.4 30.52 30.68 30.84 30.92 31.44 31.72 31.76 31.8 32 32.2 32.32 32.48 32.52 32.72 33.4 33.44 33.56 33.64 33.68 33.92 33.96 34.04 34.16 34.2 34.24 34.36 34.52 37.88 37.92 38.08 38.24 38.28 38.4 38.92 39.16 39.24 39.44 39.48 39.6 39.72 39.76 39.88 40.08 40.2 40.4 40.6 40.76 40.92 41.24 41.36 41.52 41.64 41.72 41.8 42.04 42.52 42.56 42.76 42.92 43 43.08 43.2 43.32 43.4 43.44 43.56 43.68 43.8 43.92 44.08 44.12 44.64 44.88 45 45.04 45.08 45.16 45.6 45.76 46.12 46.28 46.48 46.64 46.8 46.88 46.96 47.12 47.24 47.48 48.16 48.6 48.8 49.04 49.24 49.28 49.6 49.68 49.84 49.88 50.52 50.68 50.8 50.88 51.24 51.4 51.56 51.6 51.64 51.76 51.88 52.04 52.2 52.32 52.44 52.48 52.56 52.8 53.16 53.64 53.84 54.08 54.12 54.2 54.24 54.4 54.52 54.6 54.68 54.76 54.92 55.08 55.12 55.64 55.76 55.92 56 56.12 56.24 56.44 56.6 56.72 56.76 56.84 57 57.16 57.32 57.4 57.48 57.52 58.2 58.48 58.76 58.92 59.2 59.32 59.76 59.88 60.04 60.12 60.2 60.36 60.44 60.6 61.2 61.36 61.4 61.52 61.88 62.08 62.2 62.28 62.36 63 63.36 63.4 63.52 63.64 63.76 63.84 64 64.04 64.08 64.24 64.36 64.56 64.96 65.12 65.6 65.8 66.04 66.08 66.12 66.2 66.84 67.12 67.48 67.68 67.84 67.88 68 68.2 68.36 68.44 68.64 68.96 69.12 69.28 69.48 69.68 69.8 69.84 69.92 70.52 70.8 70.92 70.96 71.08 71.24 71.4 71.52 71.56 71.68 71.84 72 72.24 72.36 72.4 72.52 72.6 72.68 72.84 73 73.12 73.28 73.4 73.52 73.88 74 74.04 74.12 74.24 74.32 74.52 75 75.04 75.24 75.8 76.04 76.08 76.2 76.52 76.68 77.2 77.36 77.56 77.6 77.68 77.76 77.8 77.88 77.96 78.04 78.28 78.44 78.6 78.76 78.92 79.2 79.6 79.64 79.72 79.8 79.92 80.6 80.72 80.76 80.96 81 81.2 81.28 81.36 81.48 81.64 81.76 81.92 82 82.04 82.16 82.24 82.4 82.6 82.76 82.88 83.08 83.2 83.32 83.48 83.72 83.84 83.92 84 84.04 84.68 85.12 85.32 85.4 85.44 85.56 85.88 86.08 86.12 86.28 86.72 87.12 87.28 87.4 87.48 87.56 87.6 87.88 88.04 88.08 88.2 88.4 88.52 88.6 88.72 88.92 89.16 89.4 89.6 89.64 89.8 90.36 90.52 90.56 90.68 90.72 90.76 90.92 91.16 91.4 91.44 91.48 91.56 92.04 92.16 92.24 92.4 92.6 92.8 92.96 93 93.04 93.24 93.28 93.4 93.48 93.52 93.56 93.68 93.76 93.8 94 94.48 94.56 94.64 94.68 95.28 95.4 95.6 95.76 95.92 96.24 96.56 96.68 96.8 96.84 96.88 97.44 97.6 97.68 97.88 97.92 98.04 98.08 98.12 98.28 98.44 98.52 98.64 98.68 99 99.04 99.56 99.68 99.84 99.92 99.96 100.04 100.2 100.32 100.76 100.88 100.92 101 101.08 101.12 101.28 101.4 101.44 101.48 101.72 102.24 102.4 102.44 102.48 103.08 103.36 103.68 103.72 103.84 103.96 104.12 104.2 104.28 105 105.52 105.6 105.64 105.88 106.16 106.64 106.76 106.92 107.16 107.28 107.4 107.48 107.8 107.96 108.08 108.12 108.28 108.84 108.92 108.96 109.16 109.36 109.56 109.64 110.16 110.24 110.4 110.44 110.48 110.56 110.68 110.76 110.84 110.96 111.12 111.4 111.56 111.76 111.8 111.92 112 112.16 112.52 112.72 112.96 113.08 113.6 113.64 113.72 113.8 113.96 114 114.12 114.16 114.24 114.4 114.88 115 115.2 115.36 115.44 115.92 116.28 116.6 116.8 116.84 116.92 116.96 117.4 117.56 117.84 118.32 118.44 118.48 118.52 118.72 118.84 118.88 118.96 119.04 119.28 119.84 120 120.32 120.36 120.52 120.6 120.64 120.76 120.96 121.04 121.08 121.12 121.2 121.32 121.76 122.12 122.28 122.4 122.52 122.56 122.64 122.72 123.12 123.36 123.48 123.56 123.84 123.96 124.16 124.2 124.24 124.4 124.64 124.72 124.76 124.8 124.88 125.04 125.28 125.48 125.52 125.72 126.08 126.28 126.4 126.48 126.56 126.6 126.8 126.96 127.6 127.84 128.04 128.32 128.6 128.76 128.88 128.96 129.2 129.36 129.56 129.68 130.24 130.32 130.36 130.44 130.52 130.8 131 131.12 131.16 131.28 131.36 131.84 132 132.12 132.16 132.28 132.32 132.44 132.52 132.68 132.88 132.96 133.48 133.64 133.68 133.72 133.8 134 134.24 134.4 134.44 134.48 134.64 134.96 135.12 135.16 135.2 135.28 135.32 135.48 135.76 135.96 136 136.12 136.2 136.6 136.76 136.92 137 137.12 137.16 137.68 137.96 138 138.04 138.24 138.44 138.6 138.88 139 139.04 139.12 139.16 139.6 139.76 139.84 139.92 140.04 140.16 140.32 140.52 140.64 140.72 140.84 140.88 141.08 141.36 141.8 141.92 142.04 142.16 142.24 142.28 142.36 
Elapsed seconds: 188.184 s
Real time factor (RTF): 188.184 / 142.640 = 1.319
[root@david:/home/asr]# 
[root@david:/home/asr]# 

[MrJamesLZA]

lstm-transducer-small模型

验证RTF

sherpa-ncnn \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/tokens.txt \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.param \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.bin \
  ./sherpa-ncnn-lstm-transducer-small-2023-02-13/test_wavs/0.wav

执行结果(rtf=1.9)

[root@david:/home/asr]# sherpa-ncnn \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/tokens.txt \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.param \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.bin \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.param \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.bin \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.param \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.bin \
>   ./sherpa-ncnn-lstm-transducer-small-2023-02-13/test_wavs/0.wav
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-lstm-transducer-small-2023-02-13/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-lstm-transducer-small-2023-02-13/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-lstm-transducer-small-2023-02-13/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-lstm-transducer-small-2023-02-13/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-lstm-transducer-small-2023-02-13/test_wavs/0.wav
wav duration (s): 16.73
Started!
Done!
Recognition result for ./sherpa-ncnn-lstm-transducer-small-2023-02-13/test_wavs/0.wav
text:  IF UP它也是一个它这里是一个介词那它后面加的是什么呢刚才老师说的它加了是什么呢对吧它也是加 I N G对吧它为什么也就 I N G呢
timestamps: 4.04 4.32 4.64 4.84 4.96 5.08 5.16 5.72 5.84 5.96 6.2 6.32 6.4 6.6 6.68 6.8 6.92 7.08 7.2 7.32 7.44 7.52 7.6 7.64 7.72 7.96 8.04 8.16 8.28 8.36 8.48 8.64 8.76 8.84 8.96 9.04 9.08 9.12 14.2 14.24 14.36 14.44 14.56 14.72 14.92 14.96 15.04 15.2 15.24 15.6 15.72 15.84 15.88 16 16.12 16.36 16.4 16.44 16.52 
Elapsed seconds: 31.957 s
Real time factor (RTF): 31.957 / 16.730 = 1.910
[root@david:/home/asr]# 

[MrJamesLZA]

14M模型-中文

2线程跑

  sherpa-ncnn \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
    2 \
    greedy_search

结果(rtf=2.6)

 
[root@david:/home/asr]#   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
>     2 \
>     greedy_search
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=2, decoder num_threads=2, joiner num_threads=2), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)

wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.76 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.64 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 14.486 s
Real time factor (RTF): 14.486 / 5.611 = 2.581
[root@david:/home/asr]# 

4线程跑

  sherpa-ncnn \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
    4 \
    greedy_search

结果(rtf=1.9)

[root@david:/home/asr]#   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav \
>     4 \
>     greedy_search
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
wav duration (s): 5.6115
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/test_wavs/0.wav
text: 对我做了介绍那么我想说的是大家如果对我的研究感兴趣
timestamps: 0.32 0.64 0.76 0.96 1.08 1.16 1.96 2.04 2.24 2.36 2.56 2.64 2.8 3.36 3.52 3.64 3.72 3.84 3.92 4 4.08 4.24 4.48 4.56 4.72 
Elapsed seconds: 11.002 s
Real time factor (RTF): 11.002 / 5.611 = 1.961
[root@david:/home/asr]# 

用sherpa-ncnn测cctv

  sherpa-ncnn \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
    /home/asr/media/cctv-news-may15.2025/full_audio_stream.wav \
    4 \
    greedy_search

执行结果(rtf=1.9)

[root@david:/home/asr]#   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin \
>     /home/asr/media/cctv-news-may15.2025/full_audio_stream.wav \
>     4 \
>     greedy_search
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)
wav filename: /home/asr/media/cctv-news-may15.2025/full_audio_stream.wav
wav duration (s): 130.053
Started!
Done!
Recognition result for /home/asr/media/cctv-news-may15.2025/full_audio_stream.wav
text: 主播车联播今天我来说今天要说的是关税冲击下中国外贸怎么样了昨天刚刚闭幕的广交会给出了一份答案广交会一直都被视为中国外贸的风向标那从热乎乎的数据里我们可以读出很多的信息首先来看人气截至五月四号本届广交会一共吸引了来自两百一十九个国家和地区的超二十八万名境外采购商倒会叫去年同期增长了百分之十七点三这个增幅很大风雨无阻故人来这就是中国产品的吸引力我们再来看成交本节广交会现场意向出口成交是两百五十四亿美元叫去年同期增长了百分之三所以您看全球客商是在用真金白银的订单书写对中国的信心特别值得关注的是其中共建一带一路国家成交占比超过了六成是拉动成交增长的重要引擎所以有外贸商家之前在接受采访的时候说你关了一扇门我可以开无数山这就是外贸人的志气和骨气更是中国外贸实打时的底气和韧性最后我们再来看一个细节本届广交会上展出了不少融入人工智能的创新玩具吸引了全球客商纷纷下单元与此同时在美国普通的玩具却因为关税而大幅涨价所以美国挥舞关税大报最终击中的是自己当然不仅是广交会了这一段时间以来中国外贸的表现不仅我们看懂了人心里也应该有点数落外交部发言人今天介绍美方近期不断地表示希望同中方进行谈判我们的态度一直都是很明确的打分配到底谈大门敞开如果美方真想谈就应该停止威胁施压在平等尊重互惠的基础上同中方开展对话对美方而言这是唯一正确的选择
timestamps: 0 0.36 0.64 0.76 0.92 1.16 1.32 1.44 1.68 1.88 2.24 2.44 2.6 2.72 2.96 3.16 3.56 3.76 4 4.24 4.52 4.88 5.04 5.24 5.56 5.88 5.96 6.04 6.36 7.12 7.36 7.52 7.72 7.84 7.96 8.16 8.36 8.56 8.88 9.16 9.32 9.44 9.52 9.68 9.8 9.88 10.88 10.96 11.2 11.24 11.32 11.52 11.6 11.72 11.92 12.08 12.2 12.28 12.48 12.56 12.8 12.92 13.04 13.56 13.84 14.12 14.2 14.24 14.52 14.72 14.84 14.96 15.44 15.52 15.72 15.8 15.88 16.12 16.32 16.44 16.56 16.8 16.92 17.4 17.52 17.68 17.8 18.08 18.4 19.2 19.28 19.36 19.6 19.72 20.04 20.48 20.6 20.8 20.92 21.2 21.44 21.56 21.76 21.88 22.04 22.4 22.6 22.88 23.12 23.24 23.36 23.56 23.8 24.04 24.2 24.36 24.48 24.6 24.92 25.36 25.76 25.96 26.12 26.36 26.68 27 27.16 27.52 27.64 27.92 28.2 28.32 29.12 29.16 29.32 29.56 29.72 30.12 30.24 30.36 30.84 31.08 31.16 31.4 31.52 31.72 31.88 32.44 32.52 32.72 32.88 33.16 33.48 34.24 34.36 34.44 34.76 34.92 35.04 35.16 35.52 35.68 35.8 36.08 36.28 36.52 36.68 36.88 37.12 37.2 37.36 38.12 38.2 38.4 38.48 38.6 38.88 39.16 39.76 39.88 40.08 40.16 40.52 41 41.12 41.44 41.64 41.76 41.88 42.28 42.44 42.88 43.2 43.32 43.6 43.72 43.96 44.64 44.88 45.04 45.56 45.76 45.88 46.16 46.32 46.76 46.88 47 47.36 47.4 47.48 47.64 48.44 48.52 48.72 48.96 49.12 49.24 49.48 49.68 49.96 50.08 50.2 50.68 50.96 51.2 51.32 51.56 51.72 52 52.52 52.6 52.84 53.04 53.16 53.28 53.48 53.68 54.4 54.48 54.6 54.68 54.88 54.96 55.16 55.28 55.72 55.92 56.32 56.4 56.52 56.72 56.84 56.92 57.16 57.52 57.92 58.04 58.32 58.48 58.92 59.08 59.16 59.56 59.72 60.24 60.48 60.64 60.84 61 61.28 61.4 61.64 61.92 62.04 62.4 62.6 63.68 63.72 63.8 63.88 64.12 64.32 64.44 64.72 64.88 65.16 65.36 65.48 65.68 65.8 65.96 66.08 66.2 66.48 66.88 67 67.12 67.32 67.56 67.72 68.16 68.24 68.32 68.4 68.88 69.04 69.32 70.08 70.24 70.36 70.6 70.8 70.96 71.2 71.44 71.6 72 72.12 72.28 72.96 73.12 73.36 73.44 73.64 73.84 74.24 74.36 74.48 74.8 75.2 75.36 75.64 75.88 76.08 77.16 77.24 77.32 77.36 77.48 77.56 77.68 77.84 77.96 78.16 78.36 79.04 79.08 79.2 79.36 79.52 79.64 80.08 80.2 80.36 80.52 80.6 81.04 81.32 81.64 81.76 81.96 82.08 82.2 82.44 82.64 82.92 83.16 83.56 83.76 83.96 84.2 84.36 84.64 84.72 85.16 85.24 85.52 85.64 86.72 86.76 86.88 87.08 87.2 87.48 87.72 87.84 88.52 88.68 88.8 89.04 89.24 89.6 89.68 89.84 90 90.16 90.6 90.76 90.96 91.24 91.44 92.48 92.52 92.64 92.72 92.88 93.04 93.24 93.36 93.6 93.72 94.2 94.44 94.72 94.88 95.16 95.52 95.84 95.96 97.32 97.44 97.6 97.68 97.8 97.92 98.08 98.32 98.52 98.72 98.92 98.96 99.08 99.16 99.4 99.48 99.92 100.04 100.2 100.36 100.48 100.68 100.76 101.44 101.52 101.76 101.84 101.96 102.2 102.36 103.12 103.24 103.52 103.8 104.08 104.2 104.4 104.52 104.72 105.08 105.92 106 106.12 106.32 106.44 106.52 106.88 106.96 107.08 107.16 107.88 108 108.2 108.4 108.56 108.72 108.88 109.04 109.32 109.6 109.68 109.96 110.2 110.44 110.64 110.8 110.92 111.2 112.08 112.12 112.24 112.4 112.52 112.68 112.76 112.88 113 113.16 113.36 113.48 113.56 114.28 114.4 114.56 114.72 114.84 115.32 115.68 115.92 116.16 116.36 117.44 117.52 117.76 117.96 118.28 118.52 118.68 119.16 119.44 119.6 120 120.16 120.64 120.76 121.04 121.16 121.96 122.32 122.44 122.72 122.84 123.36 123.56 123.68 123.84 123.96 124.12 124.6 124.84 125 125.16 125.28 125.52 125.68 126.44 126.6 126.76 126.92 127 127.68 127.88 128.36 128.56 128.84 129.04 129.16 129.44 129.56 
Elapsed seconds: 247.600 s
Real time factor (RTF): 247.600 / 130.053 = 1.904
[root@david:/home/asr]# 

[MrJamesLZA]

20M模型-英文

2线程跑-命令

  sherpa-ncnn \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav \
    2 \
    greedy_search

执行结果(rtf=4.6)

[root@david:/home/asr]#   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav \
>     2 \
>     greedy_search
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt", encoder num_threads=2, decoder num_threads=2, joiner num_threads=2), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)

wav filename: ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav
wav duration (s): 6.625
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav
text:  AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BRAFFLELS
timestamps: 0.68 1 1.16 1.2 1.48 1.72 1.76 1.92 2 2.16 2.28 2.36 2.52 2.64 2.68 2.76 2.92 3.08 3.4 3.6 3.72 3.88 4.12 4.48 4.64 4.68 4.84 4.96 5.16 5.2 5.32 5.36 5.6 5.72 5.92 5.96 6.08 6.24 6.36 6.52 
Elapsed seconds: 31.007 s
Real time factor (RTF): 31.007 / 6.625 = 4.680
[root@david:/home/asr]# 

4线程跑-命令

  sherpa-ncnn \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin \
    ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav \
    4 \
    greedy_search

执行结果(rtf=3.5)

[root@david:/home/asr]#   sherpa-ncnn \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin \
>     ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav \
>     4 \
>     greedy_search
RecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80), model_config=ModelConfig(encoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.param", encoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/encoder_jit_trace-pnnx.ncnn.bin", decoder_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.param", decoder_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/decoder_jit_trace-pnnx.ncnn.bin", joiner_param="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.param", joiner_bin="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/joiner_jit_trace-pnnx.ncnn.bin", tokens="./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/tokens.txt", encoder num_threads=4, decoder num_threads=4, joiner num_threads=4), decoder_config=DecoderConfig(method="greedy_search", num_active_paths=4), endpoint_config=EndpointConfig(rule1=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=2.4, min_utterance_length=0), rule2=EndpointRule(must_contain_nonsilence=True, min_trailing_silence=1.4, min_utterance_length=0), rule3=EndpointRule(must_contain_nonsilence=False, min_trailing_silence=0, min_utterance_length=20)), enable_endpoint=False, hotwords_file="", hotwrods_score=1.5)



wav filename: ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav
wav duration (s): 6.625
Started!
Done!
Recognition result for ./sherpa-ncnn-streaming-zipformer-20M-2023-02-17/test_wavs/0.wav
text:  AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BRAFFLELS
timestamps: 0.68 1 1.16 1.2 1.48 1.72 1.76 1.92 2 2.16 2.28 2.36 2.52 2.64 2.68 2.76 2.92 3.08 3.4 3.6 3.72 3.88 4.12 4.48 4.64 4.68 4.84 4.96 5.16 5.2 5.32 5.36 5.6 5.72 5.92 5.96 6.08 6.24 6.36 6.52 
Elapsed seconds: 23.312 s
Real time factor (RTF): 23.312 / 6.625 = 3.519
[root@david:/home/asr]#