Skip to main content

Real-Time Container Release Notes

Important Notices​

info

It is now necessary to use processors that support Advanced Vector Extensions 2 (AVX2) when running the container in all scenarios in order to take advantage of the latest performance optimisations.

It is also recommended when using the enhanced model to use hardware that supports the AVX512_VNNI flag for optimal processing performance. The enhanced model also has increased compute requirements and will run more slowly than the standard model. For more information please see the quick start guide.

10.0.0​

Warning

This version should only be used alongside the GPU Inference Container. For more information and implementation detail, see here.

  • GPU based inference, only for English
  • Major accuracy gains for transcription (Standard and Enhanced operating points)
  • Major efficiency and speed gains, particularly for the Standard operating point
  • Improved Speaker Diarization accuracy for English (Standard and Enhanced operating points)
    • Improved numeral formatting in English
    • Improved formatting for common telephone numbers, measurements, websites, email addresses and credit cards
    • Alphanumerics now have upper-case letters
    • Added regional handling for en-AU and en-US output locale to keep 'pounds' as words
    • A number of other improvements and fixes for better readability
  • Resolved an issue where words would occasionally be fully upper-cased
  • Fix potential degradation in accuracy for multi-hour transcription sessions with low max_delay values

2.2.0​

New​

  • 14 new languages: Bashkir, Basque, Belarusian, Esperanto, Estonian, Galician, Interlingua, Marathi, Mongolian, Tamil, Thai, Uyghur, Vietnamese, and Welsh
  • The JSON-v2 output version is now 2.8, specific changes are:
    • Additional language pack information has been added to the RecognitionStarted websocket message. There is now more detailed information about properties of the language being used, such as writing direction and word delimiter.
    • We now also record the correct attachment direction for punctuation (e.g. before or after a space) in a new attaches_to field.

Improved​

  • Improved accuracy for 20 languages: Latvian (lv), Swedish (sv), Hungarian (hu), Portuguese (pt), Polish (pl), Mandarin Chinese (cmn), Arabic (ar), Dutch (nl), Slovak (sk), Bulgarian (bg), Romanian (ro), Slovenian (sl), Lithuanian (It), Croatian (hr), Malay (ms), Catalan (ca), Czech (cs), Danish (da), Greek (el), Turkish (tr)
  • Improved formatting of numeric entities such as dates, currencies and large numbers for Swedish (sv), Norwegian (no), and Dutch (nl).

Fixed​

  • Fix for accurately handling "p" as "pence" when transcribing currency in English (en).
  • Fix for handling small denominator fractions in Italian (it) and not converting to similar English homonyms e.g. "un terzo" being converted to "1/3".

Known Limitations​

Issue IDSummaryDetailed Description and Possible Workarounds
REQ-1409Proteus HCL with <unk> causes out of memory errorA custom dictionary list that contains the word <unk> causes the worker to crash.
REQ-10160Advanced punctuation for Spanish (es) does not contain inverted marks.Inverted marks [ ¿ ¡ ] are not currently available for Spanish advanced punctuation.
REQ-10627Double full stops when acronym is at the end of the sentenceIf there is an acronym at the end of the sentence, then a double full stop will be output, for example: "team G.B.."
REQ-10634Putting "-" as an item in additional vocab configuration will cause the container to failDo not enter just a "-" on its own in Custom Dictionary either as an additional vocab item or in the sounds_like property. Hyphens are still supported when entered as part of phrases or words

Supported Platforms​

Docker (17.06.0+) running on Ubuntu, Debian, Fedora or CentOS