Speech to text not able to recognize a series of 1s [234121595]

Assigned

Feature Request

Status Update

No update yet.

Description

co...@google.com

created issue #1

May 27, 2022 01:17PM

The customer is trying to transcribe tracking numbers (alphanumeric sequences) provided by their customers, spelled character by character, in French. When a sequence of consecutive “1” is spelled, it gets transcribed as a single number “1”.
They are already using the class token "$OOV_CLASS_ALPHANUMERIC_SEQUENCE".
The use of "maxAlternatives" [1] was also suggested. However, the customer stated that their system cannot handle more than one response, and it's unfeasible to implement the changes necessary to handle this option.

As such, as an example for an audio file that should have the following transcription [2], the resulting transcription is [3], the two 1s after the first 1 are not transcribed.

Would it be possible to implement this feature in the near future?

[1]

https://cloud.google.com/speech-to-text/docs/basics#synchronous-requests
[2]: "25111159"
[3]: "25159"

Comments

si...@google.com <si...@google.com> #2Nov 22, 2022 07:17AM

Assigned to gc...@google.com.

I have informed our engineering team of this feature request. There is currently no ETA for its implementation.

A current workaround would be to check the returned "boundingPoly" [1] "vertices" for the returned "textAnnotations". If the calculated rectangle's heights > widths, than your image is sideways.

[1]

https://cloud.google.com/vision/reference/rest/v1/images/annotate#boundingpoly

Issue 234121595

Description

Issue summary

Comments

si...@google.com <si...@google.com> #2Nov 22, 2022 07:17AM

Add comment

Issue metadata