Hinting
|
Name |
Values |
Description |
|---|---|---|
|
hints |
JSON-formatted string or filename |
Specifies hint phrases to be used when the ASR server converts a speech audio file into text. Hints are words or phrases for which improved recognition accuracy is desired. They typically include words not existing in the selected language model's dictionary. An example hint is "calling Verizon." Hints are captured in JSON format. Hints can be stored in a file located on the ASR server, in a file on your local computer, or provided directly in-line with a transcription request. Hints can be embedded directly in the hints field as shown in the example below.
Note that JSON uses double quotes ( Hinting is not enabled by default. If you would like to have hinting enabled, or need more information about hinting, please contact your sales representative. Refer to hints for more information on how to use file-based hinting. |
hints
Description:
Specifies hint phrases to be used when the ASR server converts a speech audio file into text. Hints are words or phrases for which improved recognition accuracy is desired. They typically include words not existing in the selected language model's dictionary. An example hint is "calling Verizon."
Hints are captured in JSON format. Hints can be stored in a file located on the ASR server, in a file on your local computer, or provided directly in-line with a transcription request.
Hints can be embedded directly in the hints field as shown in the example below.
$ curl -F hints='{"other" : ["Elastix"]}' \
-F file=@sample.wav \
-X POST http://vblaze_name:17171/transcribeNote that JSON uses double quotes (
"
) around field names and values. For this reason, use single quotes (
'
) to surround the entire JSON structure when hard coding hints in this manner.
File-based Hinting
Server-side Deployment
When hints are located within a file stored on the target ASR server, specify the name of the hint file to use in the transcription request as shown in the following example:
$ curl -F hints=sample.hints \
-F file=@sample.wav \
-X POST http://vblaze_name:17171/transcribeThe file sample.hints must exist within a language directory under /opt/voci/state/hints. For example, if the model
eng1:callcenter
is being used for transcription, the language is
eng1
. In this case, you would place sample.hints under /opt/voci/state/hints/eng1. Ensure the owner and group of this file are both set to
vocisrv
and set the file mode to
644
as shown below. If you need to create the
hints
or
eng1
directories, ensure the owner and group are set to
vocisrv
and set their modes to
755
.
$ pwd
/opt/voci/state/hints/eng1
$ ls -l
-rw-r--r--. 1 vocisrv vocisrv 128 sample.hintsClient-side Deployment
With client-side deployment, hints are stored on your local computer and provided as part of the transcription request. Store the hints in a text file located on the client computer and include the contents of this file dynamically in transcription requests. Storing hints in a file is recommended because it makes it easy to maintain and modify hints at any time with any available text editor. Managing hints client-side is recommended because it works with any number of V‑Blaze servers without having to first ensure each server has the right version of the hints file loaded.
In the example below, the
<
symbol tells curl that it should populate the
hints
field with the contents of the named file. In this example sample.hints is located on the same computer where the curl command is being executed and exists within the current working directory.
$ cat sample.hints
{
"other" : ["Elastix"]
}
$ curl -F hints="<sample.hints" \
-F file=@sample.wav \
-X POST http://voci-asr-hostname:17171/transcribe