Deploying transcription
- Kiss, Mate
Server sizing
Server sizing can be different for various speech transcription providers. When using the Verint Speech Transcription, the Verba Speech Analytics Server runs the transcription process and requires additional resources. In the case of 3rd party speech transcription providers, the transcription process runs in the cloud or on on-premise 3rd party servers. The Verba Speech Analytics Service has the following jobs (with impact):
- Execute the configured Speech Transcription policies which query the database to create a list of call records for the transcription process (low)
- Download the audio files from the storage target (low)
- Optionally transcode the audio to the format supported by the transcription provider (medium)
- Run the transcription directly on the server in the case of Verint Speech Transcription (high) or in the cloud or on 3rd party server (no impact)
- Insert plain text transcription to the database (low)
- Upload transcript file to the storage target (low)
Server Roles
The Verba Speech Analytics Service can be enabled on the following server roles:
- Speech Analytics Server
- Media Repository Server
Do not enable the service on any other server role. If the service is enabled on the Media Repository Server, the service has to be configured to limit the number of simultaneous transcription processes to 1 (Speech Analytics / Transcription / Task Processing Threads Count), this will use 1 core for transcription.
The service may be enabled on multiple servers, in this case, the system will automatically handle the load balancing between servers.
Verint Speech Transcription
When the Verint Speech Transcription engine is used, stand-alone Verba Speech Analytics Servers have to be deployed. The speech transcription process cannot be running on other server roles. This is due to the extensive resource usage nature of the speech transcription service which might interfere with other processes on the servers.
Server sizing
This table outlines typical server sizing and recommended hardware and software configurations:
Server Role | Speech Analytics Server | |
|---|---|---|
| Server Platform | Industry standard PC servers Physical or virtual | |
| CPU | Intel Xeon 2.4 GHz or higher Up to 16 CPU cores or vCPUs 2 cores/vCPUs must be "reserved" for the OS, only the rest can be used for application sizing Numbers only applicable when Receive-side scaling (RSS) is enabled in the OS | |
1 vCPU for every 200 hours of calls transcribed in 24 hours Individual results may vary due to the talk time in recordings, the storage codec, and the language model used | ||
| Memory | 4GB RAM + 2GB / language model | |
| Hard disk | System partition (OS and applications): 80 GB or more Media partition: not required, media files are only temporarily stored on the server Always use redundant disks with RAID and have separate Media and System volumes | |
| Network | All server clocks must be synchronized, typically either with the domain controller or time server All servers must have the latest Time Zone configuration Custom time zones are not supported | |
1x Gigabit Ethernet | ||
| Operating system | Microsoft Windows Server 2012 R2 Standard Edition, latest service packs installed | |
| Virtualization | VMware and Hyper-V are recommended, see Virtualization. | |
| Antivirus | Make sure your Antivirus software does not scan database, media and log folders. | |
| Power | Redundant Power Supply UPS recommended | |
3rd party Speech Transcription
In the case of 3rd party speech transcription providers, the transcription process is initiated by the Verba Speech Analytics Service, and it runs in the cloud or on on-premise 3rd party servers. For that reason, the Verba Speech Analytics Service can be run on dedicated Verba Speech Analytics Servers or on servers with Media Repository / Application Server roles.
For server sizing, use the standard Media Repository / Application Server guidelines at Server sizing and requirements.
Storage requirement
The transcription only works when the media is stored on SMB storage. If the media has to be stored on a WORM or non-SMB storage, the files should be hosted temporarily on SMB and moved to the final storage target after transcription.
Installation
Installing the Verba Speech Analytics Server
For installing the server, refer to Installing a Verba Speech Analytics Server.
Enabling the Speech Analytics Service
Step 1 - In the case of Verint Speech Transcription, copy the language model file, which is a .zip file, to C:\Program Files\Verba\resources\transcription\eliza. For additional languages, only this step needs to be repeated. The service will detect the new language model within 30 minutes, or you can force the detection by restarting the service.
Step 2 - On the Verba Web Interface, go to System / Servers. Select your server and click on the Service Activation tab.
Step 3 - Activate the Verba Speech Analytics Service by clicking on the 
icon.
Step 4 - Click on the Service Control tab.
Step 5 - Start the Verba Speech Analytics Service by clicking on the 
icon.
Creating a Data Processor
A data processor represents the connection with the transcription engine. In order to set up a speech transcription policy, a data processor has to be configured first.
Step 1 - In the Verba Web Interface go to Data / Data Processors
Step 2 - In the top right corner click on the Add new Data Processor link
Step 3 - Enter the name of the data processor
Step 4 - Select Speech Transcription for Type
Step 5 - Select the Engine
Step 6 - Click Save to save your data processor