apalii/aws-audio-recog

aws-audio-recog

Serverless audio recognition task

Resources

Steps of the implementation

audio file creation

File has been created at the https://www.narakeet.com/ with the following text from official aws docs:

Lambda runs your code on a high-availability compute infrastructure and performs all of the administration of the compute resources, including server and operating system maintenance, capacity provisioning and automatic scaling, and logging.

Cloud infrustructure entities

1. API Gateway invoke URL: https://lvydkx8cvb.execute-api.us-east-1.amazonaws.com/dev
2. S3 bucket apalii-audio-samples with audio sample
3. S3 bucket apalii-recognition-results with transribes
4. Lambda: recognition-task-consumer
5. Lambda: recognition-task-producer
6. Lambda: recognition-post-processing
7. Lambda: recognition-results
8. SQS queue: recognition
9. DynamoDB: table recognition-results
10. DynamoDB: table recognition-results

Flow

• First lambda: recognition-task-producer takes arguments from the POST request and creates task at the SQS
• SQS has another lambda recognition-task-consumer as a trigger which creates record in DynamoDB and also job at AWS Transcribe service
• AWS Transcribe creates a file with the results which will trigger the 3rd lambda recognition-post-processing
• Lambda recognition-post-processing reads the file, finds substring and saves results at the DynamoDb table recognition-results

Request example

curl -X POST \
  https://lvydkx8cvb.execute-api.us-east-1.amazonaws.com/dev \
  -H 'Connection: keep-alive' \
  -H 'Content-Type: application/json' \
  -H 'Host: lvydkx8cvb.execute-api.us-east-1.amazonaws.com' \
  -d '{
    "audio_url": "s3://apalii-audio-samples/Lambda.mp3",
    "sentences": [
        "including server and operating system",
        "can you hear me?"
    ]
}'

Results endpoint example

https://lvydkx8cvb.execute-api.us-east-1.amazonaws.com/dev/result?job_id=52fd082e-4e99-4189-9283-f420d63c5132