Я использую пример модели речевых команд из TensorFlow с переменными в Unity3d:
string WAV_INPUT = "wav_data";
string SOFTMAX_NAME = "labels_softmax";
string[] outputScoresNames = new string[] { SOFTMAX_NAME };
, а затем добавляю данные для модели следующим образом:
private void recognize(float[] audioFile)
{
//labels_softmax:0 output name
//labels wav_data:0 input name from model
string WAV_INPUT = "wav_data";
string SOFTMAX_NAME = "labels_softmax";
string[] outputScoresNames = new string[] { SOFTMAX_NAME };
int how_many_labels = 4;
string[] labels = new string[] { "_silence_" , "_unknown_", "stop","go"};
TextAsset model = Resources.Load("GoStop") as TextAsset;
TFGraph graph = new TFGraph();
graph.Import(model.bytes);
TFSession session = new TFSession(graph);
var runner = session.GetRunner();
runner.AddInput(graph[WAV_INPUT][0], audioFile);
runner.AddTarget(outputScoresNames);
runner.Run();
// float[] recurrent_tensor = runner.Run()[0].GetValue() as float[];
}
и приведенное для softmax исключение таково:
TFException: Expects arg[0] to be string but float is provided
TensorFlow.TFStatus.CheckMaybeRaise (TensorFlow.TFStatus incomingStatus, System.Boolean last) (at <1fe2de69842a4a4ba15256b83cca05f3>:0)
TensorFlow.TFSession.Run (TensorFlow.TFOutput[] inputs, TensorFlow.TFTensor[] inputValues, TensorFlow.TFOutput[] outputs, TensorFlow.TFOperation[] targetOpers, TensorFlow.TFBuffer runMetadata, TensorFlow.TFBuffer runOptions, TensorFlow.TFStatus status) (at <1fe2de69842a4a4ba15256b83cca05f3>:0)
TensorFlow.TFSession+Runner.Run (TensorFlow.TFStatus status) (at <1fe2de69842a4a4ba15256b83cca05f3>:0)
tensor.recognize (System.Single[] audioFile) (at Assets/tensor.cs:51)
tensor.Start () (at Assets/tensor.cs:23)
это проблема приведения? И как управлять ей для работы с TensorFlowSharp?