plexus.processors.DeepgramFormatProcessor module

class plexus.processors.DeepgramFormatProcessor.DeepgramFormatProcessor(**parameters)

Bases: Processor

Processor that formats Deepgram structured data into configurable text.

Reads metadata[‘deepgram’] and generates formatted text for ScoreInput.text. Unlike DeepgramInputSource (which only reads channel 0), this processor merges data from ALL channels sorted by time, properly handling stereo recordings where channel 0 = agent, channel 1 = customer.

Parameters:
format: “paragraphs” | “sentences” | “words” (default: “paragraphs”)

  • paragraphs: Groups of sentences, separated by blank lines

  • sentences: Individual sentences, one per line

  • words: Space-separated words

speaker_labels: bool (default: False) — prefix with “Speaker N: ” include_timestamps: bool (default: False) — prefix with “[12.50s]” channel: int or None (default: None) — filter to a specific channel

Example usage in YAML:
item:

class: DeepgramInputSource options:

pattern: ‘.*deepgram.*.json$’ format: paragraphs include_raw_data: true

processors:

  • class: DeepgramFormatProcessor parameters:

    format: sentences speaker_labels: true channel: 1 # Customer channel only

Initialize the processor with configuration parameters.

Args:

**parameters: Processor-specific configuration parameters

process(score_input: Score.Input) → Score.Input

Transform a Score.Input.

Args:

score_input: Input to transform (contains text, metadata, results)

Returns:

Transformed Score.Input with modified text/metadata