Hansard voice-to-text conversion and review

Use case ID: 065

Author: Bahrain Shura Council

Date: 17 October 2024
 

Objective: 

This process aims to convert spoken dialogue from weekly parliamentary meetings into accurate written texts, categorized by topic and indexed by speaker for each item of the session meeting, which the Hansard team then reviews to make the necessary corrections with minimal effort and as quickly as possible. This ensures reliable documentation of parliamentary debates for future reference and categorized by an AI model to analyze topics and historical context.

Actors: 

• Parliamentary staff: responsible for recording meetings and supervising the transcription process.
• Hansard editors: reviewing and correcting transcriptions for accuracy and clarity.
• AI model: converting audio to text, identifying the speaker, replacing colloquial words with the correct equivalent, deriving indexes by topic, by classification and by speaker, analyzing the text, linking to topics and historical reference.

Prerequisites:

• High-quality audio recording of parliamentary meetings.
• AI transcription solution capable of converting audio to text and identifying the owner of the voice.
• Access to trained Hansard editors to review transcripts.
• Integration of AI tools for text analysis and classification and dialogue.

Scenario:

1. Audio recordings of weekly Parliament meetings are captured and uploaded securely.
2. The recording is processed through intelligent transcription software, converting the spoken content into written text, identifying the MP speaking and replacing the dialect words with the correct ones.
3. The transcribed text is distributed among Hansard editors for final review.
4. The editors carefully check the text for accuracy, correct any errors or inconsistencies, and ensure clarity and fidelity to the original speech.
5. Once completed, the final template of the minutes, including its indexes and appendices, is automatically extracted and ready for approval as a first draft by the honourable members and the procedure is passed to them digitally. The text is then stored in the Hansard database for access through the website and for future research, statistical and analytical uses.
6. An artificial intelligence model is then used to analyse the completed transcripts, classify the topics discussed and identify any references to previous discussions related to these topics.

Alternative flows:

• Error Handling: If the initial transcription contains significant inaccuracies, editors may request re-transcription of specific sections to ensure reliability.

Expected results:

• Fast, accurate and reliable documentation of texts related to parliamentary meetings, which increases the efficiency of the performance of the work of the General Secretariat provided to the Mps.
• Improved decision-making, research, analysis and retrieval of information related to parliamentary discussions.
• Improved classification and contextual understanding of the topics discussed and the extraction of numerous reports and statistics that help reduce the work related to preparing annual reports on the achievements of Parliament

Potential challenges:

• Transcription accuracy: Variations in the clarity and quality of audio recordings may impact the quality of transcriptions.
• Bahraini dialect terms are numerous and varied.
• Time constraints: Tight deadlines may pressure editors to finish quickly, impacting accuracy.
• AI capabilities: The effectiveness of topic classification depends on the training of the AI model and its ability to understand context. 

Data requirements:

• High-quality audio recordings of parliamentary meetings.
• Complete, edited text documents of the transcribed sessions.
• Historical data on previously discussed topics for reference in the AI model
• Bahrain dialect vocabulary dictionary.  

Integrations with other systems:

• Integration with audio transcription software for seamless processing of meeting recordings.  
• Connection to a document management system for storing and retrieving finalized transcripts.  
• Integration with AI classification tools to enhance topic analysis and historical referencing capabilities.  
• Integration to the Parliament website
• Integration with Virtual Assistant

Success metrics:

• Accuracy of the transcriptions produced compared to spoken content.  
• Reduction in time taken to finalize and distribute transcriptions.  
• Effectiveness of the AI model in accurately classifying topics and identifying previous discussions (measured by user feedback and queries on referenced topics).  
• Frequency of retrieval and usage of classified information by parliament members and the Hansard department.
• The satisfaction of the MPs with the results of the dialogue related to the decisions and results of the sessions