speech-patho-mdl 0.1.3

Custom Trained Models for Speech Pathology

Speech Pathology Models (speech-patho-mdl)

Custom Speech Pathology Models

Quick-and-Dirty

from speech_pathology_model import ask

# On-Topic Question
answer = ask("How is velopharyngeal function typically evaluated?")
assert answer == "Velopharyngeal function during speech may also be evaluated by the measurement of pressure and airflow."

# Off-Topic Question
answer = ask("How's the Weather??")
assert not answer

The system will only return answers for on-topic questions within the scope of the knowledge base. No chit-chat or "cute" responses will be provided for off-topic or out-of-domain questions. This is the declared responsibility of the consumer.

Detailed Usage

Initialize API

api = ModelAPI()

Input

A list of tags (annotations), likely derived from unstructured text using an NLP or analytics engine

input_tags: List[str] = []

Classify Tags

d_result = api.classify(input_tags)
classification = d_result['text']

This will return a type of typedefs.dto.ServiceEvent:

class ServiceEvent(TypedDict):
    text: Optional[str]
    events: List[Dict[str, Any]]

The text attribute of this output object will be either None or have a value.

If the value is None, this means no relevant speech pathology classification was found.

If a string value does exist, this will be the top result.

The system defines the top result as

1. Having the maximum confidence level in a list of results
2. Having a confidence of at least 80%

Initialize and Invoke a Model

In the event of a classification being returned:

model = api.initialize(classification)
d_result = api.invoke(model, "How is velopharyngeal function typically evaluated?")

This result is also of type typedefs.dto.ServiceEvent and the model answer can be retrieved as

answer = d_result['text']
assert answer == "Velopharyngeal function during speech may also be evaluated by the measurement of pressure and airflow."