游雁
aad31f5e02
Revert "fix(runtime): 4 memory safety bugs in onnxruntime C++ code"
...
This reverts commit be26718559
2026-05-24 00:23:17 +08:00
游雁
eb5d0574f2
Revert "fix(runtime/go): websocket client premature close in offline mode ( #2834 )"
...
This reverts commit 13b244d4c2
2026-05-24 00:23:17 +08:00
imdenny
690280581b
fix(onnxruntime): Fix Windows MSVC compilation errors ( #2841 )
...
* fix(onnxruntime): Fix Windows MSVC compilation errors
- Fix CMake syntax warning in src/CMakeLists.txt (message function quoting)
- Replace Python-style 'and' with C++ '&&' in ct-transformer-online.cpp
- Fix ssize_t redefinition conflict between kaldi and openfst on Windows
- Update yaml-cpp CMake policies from OLD to NEW for modern CMake compatibility
These changes enable successful compilation with MSVC on Windows (VS 2022)
while maintaining compatibility with other platforms.
Tested with: Visual Studio 2022, Windows 11, CMake 3.x
* Add realtime mic app and optional FFmpeg build
2026-05-24 00:21:35 +08:00
游雁
13b244d4c2
fix(runtime/go): websocket client premature close in offline mode ( #2834 )
...
Move offline_msg_done signal after the ReadMessage loop exits (when
server closes connection), not on first message receipt. Previously,
multi-segment audio would only get the first segment transcribed
before the client closed the connection.
2026-05-24 00:18:40 +08:00
游雁
be26718559
fix(runtime): 4 memory safety bugs in onnxruntime C++ code
...
- #2848 : encoder_out_lens int32/int64 type mismatch (heap-buffer-overflow)
- #2845 : memory leak in FunTpassInferBuffer (missing delete[] for buff/len)
- #2846 : buffer leak in Audio::FfmpegLoad error paths (av_freep before avio_context_free)
- #2847 : AudioFrame copy constructor missing data/global_start/global_end (double-free risk)
2026-05-24 00:16:26 +08:00
ZZUZSL1024
a397d0b27e
Codex/create training and inference scripts ( #2791 )
...
* Update funasr_wss_server.py
* Update funasr_wss_client.py
* Add files via upload
* Update funasr_wss_server.py
* Update funasr_wss_client.py
* Add LoRA Paraformer README
2026-02-04 19:49:29 +08:00
AllenDou and zixiao
36656aa8f7
when mode is 2pass and test audio is vad_example.wav, we should give more time to websocket connection to recv message, or funasr_wss_client.py's websocket.recv() will cause exception occasionally because connection has been closed. ( #2772 )
...
Co-authored-by: zixiao <[email protected]
2026-01-07 21:00:24 +08:00
msly
c253fe7d7a
Add bounds check for msg_stamp before timestamp processing ( #2760 )
...
* Add bounds check for msg_stamp before timestamp processing
* avoids the extra layer of nesting
2026-01-01 15:12:00 +08:00
YAN XIANGJUN
995c8b090c
Fix path mistake in FunASR/runtime/python /onnxruntime/README.md and fix python dependence mistake of funasr_onnx ( #2710 )
...
* Fix path mistake in FunASR/runtime/python /onnxruntime/README.md
“FunASR/funasr/runtime/python/onnxruntime”是一个不存在的路径，在执行“cd FunASR”之后，如果用户再执行“cd funasr/runtime/python/onnxruntime”会出现错误。
"FunASR/funasr/runtime/python/onnxruntime" is an invalid path. After executing "cd FunASR", if the user execute "cd funasr/runtime/python/onnxruntime", an error will occur.
* Change onnx to jieba in the setup.py of funasr_onnx
2025-12-14 21:55:35 +08:00
Flute
fa9a6cdb1e
更新文档和运行脚本，修复文档拼写错误 ( #2688 )
...
* 更新运行脚本中的错误
现在魔搭社区已经由原来的damo转为了通义实验室的"iic"，为避免重定向，更新脚本。
* 修复文档拼写错误和更新魔搭社区路径
现在魔搭社区已经由原来的damo转为了通义实验室的"iic"，为避免重定向，更新脚本。拼写错误lagre->large
2025-10-01 14:44:28 +08:00
Gang Zhuo
6244b743a7
fix: crash when using online mode with uninitialized jieba ( #2663 )
...
Problem:
- Crash occurs with [punc_ct-transformer_cn-en-common-vocab471067-large-onnx](https://www.modelscope.cn/models/iic/punc_ct-transformer_cn-en-common-vocab471067-large-onnx ) model
- Root cause: jieba is enabled but not properly initialized
Solution:
- Add jieba initialization before usage
2025-09-20 22:41:28 +08:00
Kyle He
82a07e2f6e
fix: funasr_onnx pass run_options to ort ( #2632 )
2025-08-14 16:07:54 +08:00
nianjiuhuiyi
05c8eba11c
Fix: 修复c++后端服务因为空数组的异常退出，以及c++的http服务在收到Ctrl+C信号后无法正常退出 ( #2571 )
...
* Fix: 修复c++后端，在持续接收请求过程中运行不定时长后，服务会因为segmentation fault挂掉
添加msg_stamp是否为空的判断，在某些特定请求中，msg_stamp可能为空，避免msg_stamp为空时，直接对其进行取值操作。
* Fix: 修复c++的http服务在收到Ctrl+C信号后无法正常退出的问题
在程序主动退出时，添加工作守卫的移除，确保在捕获Ctrl+C信号后，程序能正退退出。
2025-06-27 09:57:45 +08:00
Thirteen
a3d6e48fe1
Update server.py ( #2556 )
...
Fix the error that occurs when there are no text elements in the audio and an exception is thrown.
2025-06-21 22:45:24 +08:00
lyblsgo
3ee9e8e3ba
update runtime docs
2025-06-16 19:23:20 +08:00
Logan
ab2148ec18
更新go client 的原生实现 ( #2532 )
2025-05-26 14:34:47 +08:00
王梦迪 and wangmengdi06
9038340be7
修复Fsmn_vad_online多线程调用报错 ( #2528 )
...
* 缓存cmvn file加载结果，避免多次实例化WavFrontend时重复加载
* 修复Fsmn_vad_online并发调用报错
---------
Co-authored-by: wangmengdi06 <[email protected]
2025-05-26 14:11:02 +08:00
TouchZZZ and zjw
eccf746847
fix: msg_stamp.size() is unsigned,should convert to int ( #2523 )
...
Co-authored-by: zjw <[email protected]
2025-05-22 17:47:28 +08:00
王梦迪 and wangmengdi06
fe588bc508
Fsmn_vad支持多线程并发调用 ( #2519 )
...
* 修复WavFrontend.fbank多线程调用下共享fbank_fn导致的bug
* Fsmn_vad支持多线程并发调用
---------
Co-authored-by: wangmengdi06 <[email protected]
2025-05-20 16:10:59 +08:00
zhaomingwork
ae013cf597
fix bug for core dump in http, use libboost as parse ( #2509 )
2025-05-08 23:52:09 +08:00
nianjiuhuiyi
0d6e6a2fb8
Fix: Spelling causes hotwords to not take effect ( #2444 )
...
The hotword passed by the client cannot take effect because of a spelling error.
2025-03-20 22:59:42 +08:00
辰冢
5052ca8bf0
Hotwords file needs to specify default utf-8 encoding. ( #2379 )
2025-02-12 16:13:08 +08:00
zhifu gao
3f8294b9d7
Revert "shfit to shift ( #2266 )" ( #2336 )
...
This reverts commit 1367973f98
2024-12-25 17:16:11 +08:00
Rin Arakaki
1367973f98
shfit to shift ( #2266 )
2024-12-24 17:51:31 +08:00
majic31
ae7aff2e9c
fix: solve problems in sensevoice_bin.py related to argmax and unique, as mentioned in issue #2331 ( #2332 )
2024-12-24 17:51:11 +08:00
majic31
23e7ddebcc
Fix the variable name ( #2328 )
2024-12-24 10:00:14 +08:00
shixian
a62cd7a3fd
update sensevoice onnx
2024-12-20 10:50:51 +08:00
inisis
c4412495bd
add onnxslim reference ( #2314 )
2024-12-17 12:40:04 +08:00
Haitao
7263fb08e9
识别结果中有英语时，缺少空格或者第一个单词的问题 ( #2284 )
...
* Update ct-transformer-online.cpp
修复最后两个单词之间没有空格的问题
* Update ct-transformer-online.cpp
解决语音中连续两句英语，offline结果丢失第二句第一个单词的情况。
2024-12-13 13:47:15 +08:00
莫拉古
ae49b2a8e1
变量名写错了 ( #2249 )
2024-11-29 09:55:43 +08:00
Yabin Li
141df861cb
Update websocket_protocol_zh.md
2024-11-20 15:22:59 +08:00
yijinsheng
2b747626c8
paraformer_large_offline triton运行bug 修复
2024-11-11 23:52:48 +08:00
Djraemon
7e9696f156
Fix audio format 2.0 ( #2186 )
...
* 添加了对音频文件扩展名是否为.mp3的补丁，是mp3格式则转化为wav格式
* 增加检测音频文件是否为mp3格式的补丁
* 完善对音频文件后缀名的检查，若文件后缀不是.wav，则转化为wav
* 增加音频文件后缀名检查；音频文件无效时抛出错误
* 在paraformer、vad两个模型中加入对音频文件后缀的检查，并将非wav格式转为wav格式
* 修改wav_path的数据类型，使demo能够顺利运行
2024-11-04 11:04:52 +08:00
雾聪
c540f7c831
update readme
2024-10-29 16:48:59 +08:00
雾聪
4f87e0b8f8
update readme
2024-10-29 15:06:40 +08:00
雾聪
1819303f5e
support SenseVoiceSmall in 2pass mode
2024-10-29 11:40:18 +08:00
Djraemon
a76f15c785
Fix audio format ( #2159 )
...
* 添加了对音频文件扩展名是否为.mp3的补丁，是mp3格式则转化为wav格式
* 增加检测音频文件是否为mp3格式的补丁
* 完善对音频文件后缀名的检查，若文件后缀不是.wav，则转化为wav
* 增加音频文件后缀名检查；音频文件无效时抛出错误
2024-10-21 13:30:45 +08:00
757d20b3e8
Fix typo ( #2158 )
...
* doc: Correct html5 download path error
* docs: fix typo
---------
Co-authored-by: wucan <[email protected] [email protected]
2024-10-21 13:30:04 +08:00
雾聪
1480dcf5d5
add GetInputNames GetOutputNames
2024-10-10 17:45:45 +08:00
雾聪
4d8e96f695
fix memmory leak of GetInputOutputInfo
2024-10-10 15:44:44 +08:00
Yabin Li
5b53b8fb7b
Update websocket_protocol_zh.md
2024-09-26 15:27:20 +08:00
雾聪
cd4c6485ef
update readme
2024-09-26 11:28:03 +08:00
雾聪
df24be2892
Revert "fix onnxruntime memoryleak when load model ( #2108 )"
...
This reverts commit c494d8fd18
2024-09-26 00:11:19 +08:00
雾聪
d62d237a76
add sensevoice in offline-stream
2024-09-25 23:46:47 +08:00
雾聪
aa72e0ca5f
add sensevoice-small
2024-09-25 23:44:42 +08:00
雾聪
3e44172c8b
update wbsocket for sensevoice & onnx models
2024-09-25 23:43:30 +08:00
locasxe
c494d8fd18
fix onnxruntime memoryleak when load model ( #2108 )
...
* onnxruntime memoryleak fix
* fix onnxruntime memoryleak
* fix onnxruntime memoryleak
2024-09-25 23:25:18 +08:00
lyblsgo
8745733b54
paraformer_bin bugfix
2024-09-25 20:24:42 +08:00
wusong
d20c030e5b
解决python ws服务针对尾部非人声录音无结束标识返回的问题 ( #2102 )
2024-09-25 15:11:50 +08:00
learningpro
8c88d71f34
Update readme_cn.md misspelling ( #2088 )
2024-09-20 00:42:05 +08:00
游雁