The voice-recognition-with-mxnet-and-sagemaker from cosmincatalin

voice-recognition-with-mxnet-and-sagemaker's Introduction

Voice Detection with MXNet and SageMaker

This application supports an article on my blog where I present how to build this application step by step. Full description and documentation can be found there.

Note

A run-it-on-your-own-machine version of this repository is available over here: https://github.com/cosmincatalin/dtu-voice-recognition-talk. The repo was used live in a demo at Danmarks Tekniske Universitet.

voice-recognition-with-mxnet-and-sagemaker's People

Contributors

Stargazers

Watchers

voice-recognition-with-mxnet-and-sagemaker's Issues

Sagemaker

Hey, is it possible to do the final step, so to see the results and test your data without sagemaker?
Could you explain to us what your code does:

for directory in listdir("data/test"):
    batch = []
    cnt = 0
    total = 0
    detected = 0
    for file in listdir("data/test/{}".format(directory)):
        with open("data/test/{}/{}".format(directory, file), "rb") as audio_file:
            batch.append(base64.b64encode(audio_file.read()).decode("utf-8"))
            cnt += 1
            if cnt == 5:
                binary_json = json.dumps(batch).encode("utf-8")
                response = sagemaker_runtime_client.invoke_endpoint(
                    EndpointName=predictor.endpoint,
                    Body=binary_json,
                    ContentType="application/json",
                    Accept="application/json"
                )["Body"].read()
                individual_predictions = json.loads(response, encoding="utf-8")
                for prediction in individual_predictions:
                    total += 1
                    if prediction == directory:
                        detected += 1
                cnt = 0
                batch = []
    print("""Recordings with {}:
        Total: {}
        Detected: {}
        Accuracy: {:0.2f}
    """.format(directory, str(total), str(detected), detected/total))

And how we could get the same results without sagemaker?