Metering your containers

To use Watson Speech-to-Text Embed containers in production settings, the runtime must be configured to save metering logs to persistent storage. Metering information is included in per-request log messages written to stdout.

The relevant logs messages can be identified by their JSON structure and the existence of the field total_audio_msec which is the number of milliseconds of audio processed. Below is an example log line from a request:

{"transactionId": "transId-NOTFOUND", "pl-transaction-id": "pl-transId-NOTFOUND", "x-dp-watson-tran-id": "x-dp-watson-tran-id-NOTFOUND", "userId": "userID-NOTFOUND", "bluemix-organization-guid": "orgID-NOTFOUND", "bluemix-region-id": "region-id-NOTFOUND", "bluemix-instance-id": "instance-id-NOTFOUND", "bluemix-plan-id": "plan-id-NOTFOUND", "bluemix-service-id": "service-id-NOTFOUND", "bluemix-space-guid": "space-guid-NOTFOUND", "bluemix-iamid": "iamid-NOTFOUND", "bluemix-crn": "crn-NOTFOUND", "customer_id": "customer_id-NOTFOUND", "consumer_id": "consumer_id-NOTFOUND", "customization_id": "customization_id-NOTFOUND", "acoustic_customization_id": "acoustic_customization_id-NOTFOUND", "logOptout": "0", "serviceApi": "REST", "internal": true, "audioLogFile": "/opt/ibm/chuck.x86_64/logs/2022-10/2022-10-19/payload/37a8eec318d054f84cf335d96badf00d/utt_transId-NOTFOUND_0000.wav", "resultsLogFile": "/opt/ibm/chuck.x86_64/logs/2022-10/2022-10-19/payload/37a8eec318d054f84cf335d96badf00d/result_transId-NOTFOUND_0000.log", "isCustomizationUsed": false, "isAcousticCustomizationUsed": false, "session_start_timestamp": 1666209919.1543539, "cpu_time_at_session_start": 3.03, "timenow": "2022-10-19 20:05:19.560983", "hostname": "d75420b88a09", "model_name": "en-US_Multimedia", "action": "recognize", "action_start_msec": 1666209919154, "final_response_time_msec": 406, "payload_start_msec": 104, "cpu_time_before_payload_start_msec": 80, "cpu_time_msec": 340, "request_completed": 1, "requested_features": {"maxAlternatives": 1, "wantTimestamps": true, "wantWordConfidence": false, "keywordsThreshold": null, "wordAlternativesThreshold": null, "inactivityTimeout": -1, "profanityFilter": true, "useSmartFormatting": false, "firmupSilenceTime": 0, "customizationWeight": null, "diarization": true, "profile": "low_latency", "grammarId": null, "grammarName": null, "grammarSemantics": false, "backgroundAudioSuppression": null, "speechDetectorSensitivity": null, "pciRedact": false, "semanticFirmup": false, "processingMetricsInterval": 1.0, "processingMetrics": false, "customization": false, "acoustic_customization": false, "online_grammars": false, "lowLatency": null, "audioMetrics": false, "wantInterim": false}, "deployment_info": null, "request_number": 0, "first_response_time_msec": 367, "total_audio_msec": 1220, "realtime_factor": 0.239, "busy_time_msec": 291, "utilization": 0.965, "snapshots": {"measurement_time_msec": [], "total_audio_msec": [], "cpu_time_msec": [], "busy_time_msec": [], "audio_received_msec": []}, "speech_msec": 840, "total_words_decoded": 2, "max_chunk_rtf": 0, "model_version": "en-US_Multimedia.v2021-03-03", "input_rtf": 0.005, "latency_from_eos": 0.296, "audio_confidence": 0.935}

The metered metric for Speech-to-Text is Monthly Minutes of speech audio processed. The number of minutes is rounded to a whole number per request. If the duration is less than 1 minute, round up to 1, if greater, round down to the nearest whole minute. Equivalently, divide the value of total_audio_msec by 60000, apply the floor function, and round up to 1 if the result is 0.

cat *.log | grep 'total_audio_msec' | jq --slurp '[.[] | (.total_audio_msec/60000 | floor) | if . == 0 then 1 else . end] | add'