Values: installation-dependent

Description:

The model parameter is used to specify the language model(s) to use for transcription. This parameter can be set to a single language model to transcribe all channels, or a comma-separated list of language models. V‑Cloud only supports a single language model for this parameter. Voci works with customers to ensure that their deployment delivers the best results possible, providing the language models that are most closely associated with the types of audio that each customer is transcribing. You will receive model names which are authorized for your account from Voci Support.

V‑Blaze supports a comma-separated list of models in channel order. For example, if the client is on channel 0 and the agent is on channel 1, you could use different models for each channel by setting the model parameter to model=eng1:client,eng1:agent . That setting would use the eng1:client language model to transcribe channel 0 and the eng1:agent to transcribe channel 1.

If you don't specify a value for the model parameter, the first available model of your configuration will be used. To determine the default model, use the /models API call as illustrated in the following example.

$ curl http://example:17171/models
{"models":["eng-us:callcenter","eng1:voicemail","eng1:survey"]}

As shown in the example above, if you did not specify a model when transcribing audio, the eng-us:callcenter model would be used.

Voci works with customers to ensure that their deployment delivers the best results possible, providing the language models that are best aligned with the business domain from which the speech originates.

Refer to Language models for more information on supported languages.

Values: json (default), jsontop, text, noutts

Compatibility Values: jsonlist, jsontext, jsontextlist

Description:

Specifies transcript delivery format. The following outputs are supported:

output=json

Returns a JSON output that includes utterance and event details in a hierarchical structure.

output=jsontop

Returns a JSON output in which all utterance details are collapsed into a single, top-level text field.

output=text

Returns a plain text version of the transcript.

output=noutts

Returns basic details of the transcript with utterances excluded, as in the following example:

{
    "asr": "7.1.0-1",
    "audiosecs": 6.5,
    "confidence": 0.88,
    "donedate": "2020-09-03 19:09:41.617732",
    "ended": "2020-09-03 19:09:41.619576",
    "license": "licensekey",
    "model": "eng1:callcenter",
    "nchannels": 1,
    "recvdate": "2020-09-03 19:09:41.151266",
    "recvtz": [
        "EDT",
        -14400
    ],
    "requestid": "123.456.789.101010",
    "source": "sample1.wav",
    "started": "2020-09-03 19:09:41.089314",
    "streamtags": {
        "datahdr": "WAVE"
    }
}

Tip: output=noutts can be combined with utterance_fmt for real-time applications. For example, setting output=noutts and utterance_fmt=json enables real-time utterance streaming in JSON format while excluding the utterances from post-call transcripts.

The following output parameters are legacy and can be used by manually setting output to the desired format when making a transcription request.

output=jsonlist

Returns a JSON dictionary that includes an outer list. Use to produce the same type of JSON used by previous releases. Voci recommends modifying any applications that depend on the old output formats to be compatible with the new output formats.

output=jsontext

Returns text output in the text or list formats that were previously used. A JSON representation of the transcript is provided and stored in source and utterances fields. jsontext and jsontextlist output formats are provided to enable any pre-existing applications that depend on the old output format to to be used. Voci recommends modifying any applications that depend on the old output formats to be compatible with the new output formats.

json is the recommended output format because it contains all of the value extracted from the speech signal in addition to the text of the transcript. json includes the following:

• Confidence scores for each word and utterance

• Start and stop time of each word and utterance

• Gender, emotion, sentiment, and other valuable data

In most use-cases, a transcript complete with all forms of metadata is preferable, however, text and jsontop outputs can be useful during integration testing because they are easier to read.

The unique identifier for the request for tracing purposes. This can be specified as a parameter or in the X-Request-Id HTTP header. If a requestid is provided in one of these ways, the specified requestid is included in JSON output and in the WebAPI access log.

The requestid is included in the final transcript and also in utterance callbacks as a top-level field.

The requestid can be anything. For example, you could pass an id to fetch metadata from a table or you could pass all the metadata in the requestid as a CSV string or any format you prefer, such as JSON. The following example shows a requestid :

$ curl -F "requestid=john,1234,567-uuid" -F "output=jsontop" -F "file=@/opt/voci/server/examples/sample1.wav" localhost:17171/transcribe; echo 
{"source":"sample1.wav","confidence":0.89,"donedate":"2020-01-23 13:03:02.881927","requestid":"john,1234,567-uuid","recvtz":["EST",-18000],"text":"And that it was resolved in a very professional manner. Your employees a very good.","model":"devel:callcenter","recvdate":"2020-01-23 13:03:02.276387"}

The following is an example of the utterance callback JSON with requestid included:

{"source":"sample1.wav","utterance":{"confidence":0.89,"end":6.17,"recvtz":["EST",-18000],"text":"And that it was resolved in a very professional manner. Your employees a very good.","start":0.55,"donedate":"2020-01-23 13:04:47.875351","recvdate":"2020-01-23 13:04:47.274705","metadata":{"source":"sample1.wav","model":"devel:callcenter","uttid":0,"channel":0}},"requestid":"john,1234,567-uuid"}