Endpoint with custom model returns different result to Speech Studio

Question

I have created a custom model in Speech Studio that uses sample text and structured text. I have uploaded some test samples into Speech Studio and have tested the model against these samples.

I then deployed the custom model as an endpoint and am calling this from a C# application using the C# SDK. The results I receive are different in Speech Studio compared to what I receive from the endpoint. They are considerably worse from the endpoint.

For example, I get the results below in Speech Studio:

When I run the exact same audio file through my application I get the following:

XXXX 3904 Turn heading 140

I'm not sure if it is because of the number conversion, but there are other instances where the text results are quite different.

Produces:

XXXX 184 descend to flight of all 320

The display results that I see in Speech Studio are pretty good and if I received them in my application (via the endpoint) then I would be happy.

I am currently using the code below:

var config = SpeechConfig.FromSubscription(subscriptionKey, region);

config.EndpointId = endpointId;

// Create an audio configuration

var audioConfig = AudioConfig.FromWavFileInput(file);

// Create a speech recognizer

using var recognizer = new SpeechRecognizer(config, audioConfig);

// Start recognition

var result = await recognizer.RecognizeOnceAsync();

// Check result

if (result.Reason == ResultReason.RecognizedSpeech)

{

return result.Text;

}

return null;

What can I do to get the same results as the Display Result from Speech Studio to be returned via the endpoint?