V-Blaze and V-Cloud Online Help

V‑Blaze version 7.2 (May 2021)

New Features and Improvements

  1. Additions and improvements to the V-Blaze REST WebAPI, including the following major changes. Refer to V‑Blaze REST API version 3.0.0 (May 2021) for more information on these and other changes.

    • Added several new API flows including single part post, WebSockets, and HTTP results streaming.

    • Added new stats endpoint, and made updates to the status, statusfull, and sysinfo endpoints.

    • Made multiple language ID (LID) improvements and additions. Standard LID behavior is to now score and make decisions across an entire call, independent of the number of channels in the audio. This will result in the JSON element lidinfo always being a single dictionary rather than a list.

  2. Added several new vociserver service commands: sysinfo, stats, status, and params. Refer to V‑Blaze Service Commands for more information.

  3. Added soft shutdown feature for the V-Blaze REST API. When the REST API is shut down, V‑Blaze now allows active streams a fixed amount of time to complete and prevents any additional streams from being submitted. If an active stream does not complete in the allotted time, the stream will be terminated and the results will be lost. In versions prior to V‑Blaze 7.2, any active streams are terminated upon shutdown of the API.

    Configure the time to wait for active streams to complete after a shutdown request by setting the stoptimeout option in /opt/voci/state/webapi.cfg to the number of seconds to wait. By default, stoptimeout has a value of 60 seconds. Note that stoptimeout has an effective upper bound of 5 minutes. Refer to V‑Blaze API Configuration (Optional) for more information on the webapi configuration file.

    When using the REST API service script, executing a vociwebapi stop or a vociwebapi restart will perform a soft stop by default. If an immediate hard stop is needed for any reason, a vociwebapi stop —force or a vociwebapi restart —force may be used skip or interrupt the soft stop process.

  4. The ASR Engine now loads acoustic models in a way that optimizes GPU memory usage while avoiding out of memory issues that could prevent service initialization. Refer to Language Model Configuration for more information.

  5. Made multiple improvements to number translation and web URL formation:

    • Changed number translation behavior to improve transcript readability. The translation is now more conservative by considering more context for various situations. More control over concatenation of numbers is now available with the addition of the numtrans=concat option. Refer to numtrans for more information on this parameter.

  6. Removed limit of number of concurrent audio streams that could be processed.

  7. Made multiple improvements to redaction functionality:

    • Added the scruboffset option when using redaction. Refer to Redaction for more information on this parameter.

    • A recommended/default configuration for audio and text scrubbing is now distributed as a packaged dependency. This configuration is located at /opt/voci/scrub.conf. Existing customized configurations located in /opt/voci/state/scrub.conf will override the distributed configuration. Previously, if no scrub.conf was installed on a system, nominal behavior was to redact all digits (0-9) without exception. The recommended configuration makes exceptions for numeric elements such as ordinals (1st, 2nd, etc.), clock times, and prices. This configuration was previously recommended by Voci support and installed manually on most systems as /opt/voci/state/scrub.conf. Customers who wish to always use Voci recommendations can now simply remove the redundant /opt/voci/state/scrub.conf.

    • Fixed a rare truncated audio issue when using redaction.

  8. Updated licensing to make local maxstreams configuration setting behavior consistent with maxdecoders. The license server can specify a maximum; otherwise, the system default or locally configured value is used.

  9. Improved error handling:

    • Fixed an issue that could result in ASR errors not being reported correctly.

    • JSON output now only includes the ended, model, nchannels, and audiosecs elements if the stream completed successfully. These elements do not display in JSON output if there was a problem processing audio.

  10. Added complete support for new language model naming formats. For example, ita-it:callcenter, which corresponds to the language-dialect:domain.

  11. Minor bug fixes and general improvements.