medkit.audio.transcription

medkit.audio.transcription#

Submodules#

Classes#

`DocTranscriber`	Speech-to-text transcriber generating text documents from audio documents.
`TranscriptionOperation`	Protocol for speech-to-text transcription operations.
`TranscribedTextDocument`	Text document generated by audio transcription.

Package Contents#

class medkit.audio.transcription.DocTranscriber(input_label: str, output_label: str, transcription_operation: TranscriptionOperation, attrs_to_copy: list[str] | None = None, uid: str | None = None)#

Bases: medkit.core.Operation

Speech-to-text transcriber generating text documents from audio documents.

For each text document, all audio segments with a specific label are converted into text segments and regrouped in a corresponding new text document. The text of each segment is concatenated to form the full raw text of the new document.

Generated text documents are instances of TranscribedTextDocument (subclass of TextDocument) with additional info such as the identifier of the original audio document and a mapping between audio spans and text spans.

Methods :func: create_text_segment() and :func: augment_full_text_for_next_segment() can be overridden to customize how the text segments are created and how they are concatenated to form the full text.

The actual transcription task is delegated to a TranscriptionOperation that must be provided, for instance :class`~medkit.audio.transcription.hf_transcriber.HFTranscriber` or :class`~medkit.audio.transcription.sb_transcriber.SBTranscriber`.

Parameters:

input_label: str: Label of audio segments that should be transcribed.
output_label: str: Label of generated text segments.
transcription_operation: TranscriptionOperation: Transcription operation in charge of actually transcribing each audio segment.
attrs_to_copy: list of str, optional: Labels of attributes that should be copied from the original audio segments to the transcribed text segments.
uid: str, optional: Identifier of the transcriber.

init_args#

input_label#

output_label#

transcription_operation#

attrs_to_copy#

_attr_label#

run(audio_docs: list[medkit.core.audio.AudioDocument]) → list[medkit.audio.transcription.transcribed_text_document.TranscribedTextDocument]#

Return a transcribed text document for each document in audio_docs.

Parameters:

audio_docs: list of AudioDocument: Audio documents to transcribe

Returns:

list of TranscribedTextDocument:: Transcribed text documents (once per document in audio_docs)

_transcribe_doc(audio_doc: medkit.core.audio.AudioDocument) → medkit.audio.transcription.transcribed_text_document.TranscribedTextDocument#

augment_full_text_for_next_segment(full_text: str, segment_text: str, audio_segment: medkit.core.audio.Segment) → str#

Append intermediate joining text to full text before the next segment is concatenated to it.

Override for custom behavior.

class medkit.audio.transcription.TranscriptionOperation#

Bases: typing_extensions.Protocol

Protocol for speech-to-text transcription operations.

Attributes:

output_labelstr: Label to use for generated transcription attributes.

output_label: str#

run(segments: list[medkit.core.audio.Segment])#

Run the transcription operation.

Add a transcription attribute to each segment with a text value containing the transcribed text.

Parameters:

segments: list of AudioSegment: List of segments to transcribe

class medkit.audio.transcription.TranscribedTextDocument(text: str, text_spans_to_audio_spans: dict[medkit.core.text.Span, medkit.core.audio.Span], audio_doc_id: str | None, anns: Sequence[medkit.core.text.TextAnnotation] | None = None, attrs: Sequence[medkit.core.Attribute] | None = None, metadata: dict[str, Any] | None = None, uid: str | None = None)#

Bases: medkit.core.text.TextDocument

Text document generated by audio transcription.

Parameters:

text: str: The full transcribed text.
text_spans_to_audio_spans: dict of TextSpan to AudioSpan: Mapping between text characters spans in this document and corresponding audio spans in the original audio.
audio_doc_id: str, optional: Identifier for the original AudioDocument that was transcribed, if known.
anns: sequence of TextAnnotation, optional: Annotations of the document.
attrs: sequence of Attribute, optional: Attributes of the document.
metadata: dict of str to Any: Document metadata.
uid: str, optional: Document identifier.

Attributes:

raw_segment: TextSegment: Auto-generated segment containing the raw full transcribed text.

text_spans_to_audio_spans: dict[medkit.core.text.Span, medkit.core.audio.Span]#

audio_doc_id: str | None#

get_containing_audio_spans(text_ann_spans: list[medkit.core.text.AnySpan]) → list[medkit.core.audio.Span]#

Return the audio spans used to transcribe the text referenced by a text annotation.

For instance, if the audio ranging from 1.0 to 20.0 seconds is transcribed to some text ranging from character 10 to 56 in the transcribed document, and then a text annotation is created referencing the span 15 to 25, then the containing audio span will be the one ranging from 1.0 to 20.0 seconds.

Note that some text annotations maybe be contained in more than one audio spans.

Parameters:

text_ann_spans: list of AnyTextSpan: Text spans of a text annotation referencing some characters in the transcribed document.

Returns:

list of AudioSpan: Audio spans used to transcribe the text referenced by the spans of text_ann.

to_dict(with_anns: bool = True) → dict[str, Any]#

classmethod from_dict(doc_dict: dict[str, Any]) → typing_extensions.Self#

Create a TranscribedTextDocument from a dict.

Parameters:

doc_dict: dict of str to Any: A dictionary from a serialized TranscribedTextDocument as generated by to_dict()