This is a guide for customers who are updating to V8.0.0 or later (October 2020). It documents changes in the batch container, and how you, a customer, may need to reintegrate your batch container with any other systems. It is provided in addition to our standard release notes and documentation pack.
As part of this upgrade, some V1 features that are no longer supported have been completely deprecated, and will cease to work as announced in the v6.2.0 release.
In all cases, replacements are supported via our V2 input, and are documented in our Speech API Guide.
The changes below should show no loss of any feature or functionality as a result of the migration.
The scope of this document shows:
The scope of this document excludes
Previously Speechmatics built batch containers with their own license file integrated within the container for each language required by a customer. For simplicity and replicability we have moved to a generic customer-agnostic container for each language, with each customer now receiving a separate license file to use with the container(s) they are licensed for.
Please note: The contents of the license file are confidential. They should be shared on the principles of least privilege. Speechmatics is not responsible for how you handle, store, or share licensing information.
Speechmatics Support will provide you with a new license file. The license is a JSON file called license.json and has the following JSON structure:
| Item | Description |
|---|---|
| Customer name | This is your company's name |
| Id | This is internal to Speechmatics |
| Is-Trial | Whether the license is for a trial use of Speechmatics or not |
| Metadata | What Features a container is licensed to use. These can include: Speaker Diarization Channel Diarization Speaker Change Batch Container use Real-time container use Language: any supported language Language: A supported individual language (e.g. English) |
| NotValidAfter | The date after which the license expires and can no longer be used to run the container. The date is in ISO format |
| ValidFrom | The date from which this license is valid. |
| Signed Claims Token | A unique reference number used to validate the license file when running the container. Generated by Speechmatics |
The values in this license file will reflect each customer's individual contract arrangement with Speechmatics.
An example license file is below:
{
"contractid": 1,
"creationdate": "2020-03-24 17:43:35",
"customer": "Speechmatics",
"id": "c18a4eb990b143agadeb384cbj7b04c3",
"is_trial": true,
"metadata": {
"key_pair_id": 1,
"request": {
"customer": "Speechmatics",
"features": [
"MAPBA",
"LANY"
],
"isTrial": true,
"notValidAfter": "2021-01-01",
"validFrom": "2020-01-01"
}
},
"signedclaimstoken": "example",
}
Previously the batch container was licensed by use of the environment variable LICENSE_KEY. This is no longer a valid variable and will not license the product. Instead you may either license the product via the two methods described below:
docker run -i -v $AUDIO_FILE:/input.audio -v $CONFIG_JSON:/config.json -v /my_license.json:/license.json batch-asr-transcriber-en:8.0.0
LICENSE_TOKEN variable when running a transcription job. See an example of using LICENSE_TOKEN below:docker run -i -v $AUDIO_FILE:/input.audio -v $CONFIG_JSON:/config.json -e LICENSE_TOKEN='example' batch-asr-transcriber-en:8.0.0
If you lose a license file or it is no longer secure, Speechmatics can generate a new one. Please contact Speechmatics support if this is the case.
In the Speechmatics container you can still process a media file for transcription without use of the V2 configuration object. This will generate our JSON v2 output without any alteration or changes to the text.
From the V8.0.0 release, the configuration file is now the only way by which you can modify the transcription output in the Speechmatics container. If you want to use features such as diarization, punctuation overrides, output locale etc. you must use the configuration object to request these features.
If you already do so, then you do not need to make any changes to how you use the container.
All JSON transcription output will now be in the V2.4 output.
As part of the v7.0.0 release support for V1 features was withdrawn. As part of this release all V1 features have now since been removed. Where applicable, these have been replaced by options within the configuration object. This includes the following:
| V1 Item | Type | Replaced By |
|---|---|---|
| DIARIZE. Enables speaker diarization | environment variable | Use the diarization:speaker parameter within the configuration object |
| DIARISE. Enables speaker diarization | environment variable | Use the diarization:speaker parameter within the configuration object |
| CHANNEL_DIARISATION. enables channel diarization on stereo files | environment variable | Use the diarization:channel parameter within the configuration object |
| CHANNEL_DIARISATION_LABELS. Provides labels to different speakers when using channel diarization | environemnt variable | Replaced by the parameter channel_diarization_labels in the configuration object |
| LICENSE_KEY. used to license the batch container | environment variable | Replaced by LICENSE_TOKEN |
/extra_words.txt. Used as a custom dictionary to generate additional vocabulary objects | text file | Use the additional vocab parameter within the configuration object to generate a custom dictionary |
/build_date. Documents the date the batch container was built by | text file | Replaced by the new licensing file, and no longer needed |
/license_days. How many days the license has to run | text file | Replaced by the new licensing file, and no longer needed |
Notifications are still supported in the batch container as before. There are a few changes in how single and multi-part notifications are generated and encoded, and this is noted below for integration purposes:
transcript, this will now be output in the JSON-V2 format rather than the deprecated V1 JSON formatcontents to be blank by using []. An example is provided belowcharset=utf8 on all transcript types. Ensure that your workflow can support thisContent-Type header's used to be set always to application/octet-stream . This value now corresponds to actual content of the notification and is application/json in case of JSON-v2 content, text/plain in case of an SRT contetn, and application/octet-stream for TXT contentAn example notification configuration that would generate a notification with no contents is shown below. This is a change from the previous version of batch container.
{
"notification_config": [{
"url": "http://localhost:8080",
"contents": []
}]
}