Changelog

September 2022

Transcription Language Skill

• The Transcription Skill converts audio to text, which can be then analyzed and transformed by other Language Skills
• Labels extracted from transcriptions hold timestamps of the labeled words in the audio input
• Automated diarization- different speakers are identified and exchanges are separated into different sections
• The pipeline API supports .wav & .mp3 file inputs. The Transcription Skill is required to process audio files.
• Use the async endpoint (details below) to process big audio files asynchronously

```const pipeline = new oneai.Pipeline(
   oneai.skills.transcribe(),
   oneai.skills.emotions(),
   oneai.skills.summarize(),
);

const output = await pipeline.runFile('my-conversation.mp3');
console.log(output.transcription.text);                  // transcribed text
console.log(output.transcription.emotions);        // emotions that appear in the conversation
console.log(output.transcription.summary.text); // summary of the conversation```

New & Improved Language Skills

Headline Generation

• Generate an appropriate headline for your input text.

Subheadings Generation

• Generate a subheading for your input text.

Summarize

• The Summary Skill now generates `origin` labels, mapping words in the summary to their position in the source input (string indices for text inputs, timestamps for audio inputs).

Split-by-Topic

• Skill is now configurable, providing control over how many sections the input is split into.
• Combine with the Headline / Subheadings Skills to generate headline per section

Numbers

• Numbers and quantities that appear in text form are interpreted & provided as numeric data
• Dates, times & time-durations in various formats are provided in a standardized date string
• Added a `datetime_baseline` parameter to control the current time dates & times should use as a baseline

Names

• The Names Skill detects when names that appear in text reference real-world entities (people, companies, locations, etc.) based on context.
• `name` labels are enriched with entity data when known entities are recognized.

Language-Detection

• Automatically identify the dominant language used in the input and help with further processing and translation.

Analytics API (beta)

• The analytics engine makes large amounts of text data digestible, by clustering together texts with similar meaning and accumulating metadata generated by Language Skills (such as Sentiment & Topics).
• The API accepts text items and organizes them in hierarchical clusters by meaning.
• Generated clusters can be fetched from the API or reviewed at the Analytics section of the Studio (the UI is open-source)
• Cluster collections can be queried with specific text items, fetching the cluster with the most similar meaning, making the analytics API an effective tool for intent based classification and search.
• Text items can be enriched with metadata generated by Language Skills. Clusters then aggregate item metadata to derive insights at scale.
• Quickstart guide - docs.oneai.com/docs/quick-start-analytics

Async API

• The async API endpoints introduce asynchronous processing of large text inputs and audio files.
• The upload request of the input and the response with the output are split into separate endpoints, so outputs can be retrieved without waiting for the entire processing time.
• Use the `/async`endpoint to process raw text
• Use the `/async/files` endpoint to process binary encoded files
• Successful async requests will return a `task_id` parameter that was assigned to the input.
• Task status and outputs can be fetched via polling from the `/async/tasks` endpoint

‍

‍