ctc_decoder icon indicating copy to clipboard operation
ctc_decoder copied to clipboard

A ctc decoder for both online and offline asr model

Results 8 ctc_decoder issues
Sort by recently updated
recently updated
newest added

你好,请问有使用语言模型进行联合解码的脚本吗。我是从wenet那边过来的,我正尝试使用triton部署离线语音识别服务,在不含语言模型的情况下可以跑通并且获取正确结果,但是加上语言模型之后无法获取正确结果。 服务端显示内容如下: E0524 08:34:07.612731 97 python.cc:1968] Stub process is unhealthy and it will be restarted. Loading the LM will be faster if you build a binary file. Reading /ws/onnx_model/lm.arpa ----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100...

你您好,请问这个在windows可以安装吗

现在如果arpa文件中,有中英混合的情况,输出的score_hyps都是诸如下面的情况: (((-3.4028234663852886e+38, ()), (-3.4028234663852886e+38, (1819, 29)), (-3.4028234663852886e+38, (29,)), (-3.4028234663852886e+38, (1819,)), (-3.4028234663852886e+38, (1819, 29, 2327)), (-3.4028234663852886e+38, (1819, 2327)), (-3.4028234663852886e+38, (29, 2327)), (-3.4028234663852886e+38, (2327,)), (-3.4028234663852886e+38, (1819, 5170)), (-3.4028234663852886e+38, (5170,))), ((-3.4028234663852886e+38, ()), (-3.4028234663852886e+38,...

大佬,我看见你提的这个[issues]( https://github.com/NVIDIA/NeMo/issues/215),我想请教一下,英文语音识别使用bpe建模,语言模型训练数据是怎么个形式呢?如果是如下这种形式 THANK YOU _THANK _YOU 能使用这个ctc_beam_search_decoder_batch函数解码吗

In this pr,fix language model score repeatedly. When hotwords_scorer->is_character_based and ext_scorer->is_character_based() is false,The language model and hot word scores will be repeatedly calculated. In fact, if the language model is...

``` git clone https://github.com/Slyne/ctc_decoder.git && cd ctc_decoder/swig && bash setup.sh ``` ``` python3 -c "import swig_decoders" ``` ModuleNotFoundError: No module named '_swig_decoders'

如果语言模型是分词的,is_character_based_==False, 分词语言模型只有遇到space_id才打分 make_ngram[4]得到截止当前帧固定长度窗口内的词语。 如果输出token没有SPACE_ID_,只取当前帧固定长度窗口内的词语,前面的词不就没有取到的机会吗? ``` // language model scoring float ngram_score = 0.0; if (ext_scorer != nullptr ) { if (hotwords_scorer != nullptr && !hotwords_scorer->hotwords_dict.empty() && !(hotwords_scorer->is_character_based ^ ext_scorer->is_character_based()) &&...