Does sherpa-onnx work on AVX processor

[mercurial-moon]

Hi would sherpa-onnx work on a processor that doesn't have AVX2 instruction?
Reason why i'm asking is when I run a sample program on a PC with only AVX instruction it failed silently.

The relevant code is

           var config = new OnlineRecognizerConfig();
            config.FeatConfig.SampleRate = 16000;
            config.FeatConfig.FeatureDim = 80; // change this for parakeet
            config.ModelConfig.Transducer.Encoder = @"..\sherpa_onnx_nemo_parakeet_tdt_0.6b_v2\encoder.int8.onnx";
            config.ModelConfig.Transducer.Decoder = @"..\sherpa_onnx_nemo_parakeet_tdt_0.6b_v2\decoder.onnx";
            config.ModelConfig.Transducer.Joiner = @"..\sherpa_onnx_nemo_parakeet_tdt_0.6b_v2\joiner.onnx";


            config.ModelConfig.Paraformer.Encoder = null;
            config.ModelConfig.Paraformer.Decoder = null;

            config.ModelConfig.Tokens = @"..\sherpa_onnx_nemo_parakeet_tdt_0.6b_v2\tokens.txt";
            config.ModelConfig.Provider = "cpu";
            config.ModelConfig.NumThreads = 1;
            config.ModelConfig.Debug = 1;

            config.DecodingMethod = "greedy_search"; // or modified_beam_search
            config.MaxActivePaths = 4; // only when decoding method is modified_beam_search
            config.EnableEndpoint = 1;

            config.Rule1MinTrailingSilence = 2.4f;
            config.Rule2MinTrailingSilence = 0.8f;
            config.Rule3MinUtteranceLength = 20.0f;

            recAi = new OnlineRecognizer(config); // !!<- at this line the memory increases to 1.6GB and the code stops running in Visual Studio 
            recAiStream = recAi.CreateStream();

At the indicated line in the above code the memory starts increasing and flattens out at 1.6GB and then the Visual Studio comes out of debugging mode.

Also exit code as under ( No other errors shown in output panel)

program has exited with code 4294967295 (0xffffffff).

[csukuangfj]

It depends on onnxruntime. Are you able to run any model using onnxruntime, not necessarily using sherpa-onnx.

[mercurial-moon]

Trying out onnx for the first time, I used to use faster whisper / whisper models.
So probably i overlooked some thing.
I only installed sherpa.onnx 1.12.6 from nuget. I'm running on Win 11.
How do I test to run any model on onnxruntime?

[mercurial-moon]

I did another test, but this time in Release mode and on a pc with AVX2 instructions.
This time there was an immediate crash on trying to Record from mic.
The important portions of the stack trace are

.NET Version: 8.0.13
Description: The process was terminated due to an unhandled exception.
Exception Info: System.AccessViolationException: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
Stack:
   at SherpaOnnx.OnlineRecognizer.SherpaOnnxCreateOnlineStream(IntPtr)
   at SherpaOnnx.OnlineRecognizer.SherpaOnnxCreateOnlineStream(IntPtr)
   at SherpaOnnx.OnlineRecognizer.CreateStream()
   at SherpaOnnxTest.Form1.Record() 

I found the errors in Event Viewer

[csukuangfj]

You need to.run it inside a terminal and post all logs.

[mercurial-moon]

Hi, there are no logs since it fails silently and its a Windows Forms App the only logs i could gather with a release build was to look inside Event Viewer

[csukuangfj]

Before you write any code of your own, please first try our examples.

[csukuangfj]

Before you write any code of your own, please first try our examples.

You did not follow this comment. @mercurial-moon

[csukuangfj]

No informative logs, no help.

[mercurial-moon]

No informative logs, no help.

please let me know where do I find them, or what information do you need. I'm willing to provide them... It's just that i can't locate the logs.

[csukuangfj]

1. Don't write any code before you get our examples working
2. Run our example without a microphone from the terminal, i.e., the one that decodes a file
3. Run our example with a microphone from the terminal
4. If everything works, you can look at the code of our example and begin to write your own.