Emo model onnx improvements

[leoromanovich]

Hi, @georgygospodinov!

I've tried to convert to onnx and run Emo model, but found the following problems which I've fixed in this PR.

• strange representation of 1d avg pool + linear layer in onnx graph was found so inference for that graph is not working because AvgPool -> Gemm array shapes mismatch (attached pics)
• no onnx inference for Emo model was found too

to address these problems I'm offering the following changes:

• replace avg pool with kernel_size == -1 dim to simple toch.mean over this dimention (seem's like, in that context operations are equivalent and result is not changing)
• add recognise_emotion method to infer onnx version of emo model
• add example of emo model inference to inference_example.ipynb

@georgygospodinov , may I ask you to review my PR?
I would appreciate any comments, ty!

Minimal code sample to reproduce graph problem:

import torch
import onnxruntime as rt

import gigaam
from gigaam.onnx_utils import DTYPE, FEAT_IN, SAMPLE_RATE

if __name__ == "__main__":

    opts = rt.SessionOptions()
    opts.intra_op_num_threads = 16
    opts.execution_mode = rt.ExecutionMode.ORT_SEQUENTIAL

    session = rt.InferenceSession(
        "path_to_onnx_checkpoint/v1_emo.onnx", 
        providers=["CPUExecutionProvider"],
        sess_options=opts,
        )

    preprocessor = gigaam.preprocess.FeatureExtractor(SAMPLE_RATE, FEAT_IN)

    input_signal = gigaam.load_audio(str("path_to_example_file/example.wav")).unsqueeze(0)
    input_signal = preprocessor(
        input_signal,
        torch.tensor([input_signal.shape[-1]]),
        )[0].numpy()

    enc_inputs = {
        node.name: data
        for (node, data) in zip(
            session.get_inputs(),
            [input_signal.astype(DTYPE), [input_signal.shape[-1]]],
            )
        }

    enc_features = session.run(
        [node.name for node in session.get_outputs()], enc_inputs
        )[0]

    print(enc_features)

Output:

2024-12-20 13:50:00.905064261 [E:onnxruntime:, sequential_executor.cc:514 ExecuteKernel] Non-zero status code returned while running Gemm node. Name:'/head/Gemm' Status Message: /onnxruntime_src/onnxruntime/core/providers/cpu/math/gemm_helper.h:17 onnxruntime::GemmHelper::GemmHelper(const onnxruntime::TensorShape&, bool, const onnxruntime::TensorShape&, bool, const onnxruntime::TensorShape&) left.NumDimensions() == 2 || left.NumDimensions() == 1 was false.

Traceback (most recent call last):
  File "/mnt/original_repo/GigaAM/infer_model.py", line 35, in <module>
    enc_features = session.run(
                   ^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 220, in run
    return self._sess.run(output_names, input_feed, run_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Gemm node. Name:'/head/Gemm' Status Message: /onnxruntime_src/onnxruntime/core/providers/cpu/math/gemm_helper.h:17 onnxruntime::GemmHelper::GemmHelper(const onnxruntime::TensorShape&, bool, const onnxruntime::TensorShape&, bool, const onnxruntime::TensorShape&) left.NumDimensions() == 2 || left.NumDimensions() == 1 was false.