Azure - Speaker Diarization Benchmark

Prerequisites

• Ubuntu 20.04 (x86_64)
• Git
• Python 3.7+
• PIP
• Azure Account

Usage

1. Clone the repository:

git clone https://github.com/Picovoice/speaker-diarization-benchmark.git

2. Install the dependencies:

pip3 install -r requirements.txt

3. Set up the dataset as described in the main readme of the repository.

4. A client library for the Speech to Text REST API should be generated, as outlined in the documentation.

5. Create an Azure Storage account on you Azure account.

6. Run the benchmark:

python3 benchmark.py \
    --type {ACCURACY|CPU|MEMORY} \
    --dataset ${DATASET} \
    --data-folder ${DATA_FOLDER} \
    --label-folder ${LABEL_FOLDER} \
    --engine AZURE_SPEECH_TO_TEXT \
    --azure-storage-account-name ${AZURE_STORAGE_ACCOUNT_NAME} \
    --azure-storage-account-key ${AZURE_STORAGE_ACCOUNT_KEY} \
    --azure-storage-container-name ${AZURE_STORAGE_CONTAINER_NAME} \
    --azure-subscription-key ${AZURE_SUBSCRIPTION_KEY} \
    --azure-region ${AZURE_REGION}

Where:

• type is the type of benchmark to run. It can be ACCURACY, CPU, or MEMORY.
• dataset is the name of the dataset to use.
• data-folder is the path to the folder containing the audio files.
• label-folder is the path to the folder containing the ground truth labels.
• engine is the name of the engine to benchmark. It must be AZURE_SPEECH_TO_TEXT.
• azure-storage-account-name is the name of the Azure Storage account to use.
• azure-storage-account-key is the key of the Azure Storage account to use.
• azure-storage-container-name is the name of the Azure Storage container to use.
• azure-subscription-key is the subscription key to use.
• azure-region is the region to use.