Deploying Hugging Face (PyAnnote) Speaker Diarization Model for Asynchronous Processing on Amazon SageMaker

In this article, we will walk through the steps of deploying a Hugging Face Speaker Diarization model for asynchronous processing on Amazon SageMaker. Asynchronous processing allows you to initiate a prediction and retrieve the results later, which can be beneficial when processing large volumes of data or when the predictions are computationally intensive.

Prerequisites

To follow this tutorial, you will need the following:

• An Amazon SageMaker account
• The Hugging Face PyAnnote library
• A trained speaker diarization model (e.g., the pyannote/speaker-diarization-avg-s model)

Creating the Endpoint Config

1. Open the Amazon SageMaker console and click on Create endpoint config.
2. Select Create new endpoint config and give it a name.
3. For the `Model` field, select Hugging Face and then select your trained speaker diarization model.
4. For the `Accelerator type` field, select ml.c5.4xlarge.
5. For the `Instance type` field, select ml.c5.4xlarge.
6. For the `Endpoint config name` field, give it a name.
7. Click on Create endpoint config.

Creating the Endpoint

1. Open the Amazon SageMaker console and click on Create endpoint.
2. Select Create new endpoint and give it a name.
3. For the `Endpoint config` field, select the endpoint config you created in the previous step.
4. For the `Async invocation` field, select Enabled.
5. For the `Endpoint name` field, give it a name.
6. Click on Create endpoint.

Using the Endpoint

Once the endpoint is created, you can use it to perform speaker diarization on audio files. To do this, you can use the following Python code:

“`python
import boto3

# Create a SageMaker client
sagemaker_client = boto3.client(sagemaker)

# Endpoint name
endpoint_name = YOUR_ENDPOINT_NAME

# Input audio data
with open(input.wav, rb) as f:
audio_data = f.read()

# Send the audio data to the endpoint
response = sagemaker_client.invoke_endpoint(
EndpointName=endpoint_name,
Body=audio_data,
)

# Get the prediction results
result = response[Body].read().decode(utf-8)

# Print the prediction
print(result)
“`

Conclusion

In this article, we walked through the steps of deploying a Hugging Face Speaker Diarization model for asynchronous processing on Amazon SageMaker. This allows you to process large volumes of audio data efficiently and retrieve the results when they are ready.

Kind regards,

J.O. Schneppat.

Commercial