Semi-automated production of verbatim reports of sittings using automatic speech recognition

Use case ID: 002

Author: IT Department, Chamber of Deputies of Italy

Date: 22 May 2024

Objective: 

Produce draft verbatim reports of parliamentary sittings using solutions based on automatic speech recognition (ASR). The draft is reviewed by humans and then published.

Actors: 

• Business units
• ASR system

Prerequisites:

• Existence of audio signal of speeches 
• Existence of a pre-trained speech-to-text model

Scenario:

1. A new segment of audio signal of a speech is available (e.g. one five-minute segment).
2. The AI speech-to-text model produces the corresponding text.
3. The text is reviewed by colleagues.
4. All the text segments are concatenated and reviewed.
5. The final verbatim report is published.

Expected results:

• The process of producing verbatim reports of sittings is more efficient.

Potential challenges:

• ASR model not properly trained
• The ASR model might not work well with dialects and specific terms not used during the training phase (e.g. the term “COVID-19” was used a lot during the pandemic period, but it was unknown before)

Data requirements:

• The ASR model should be trained on a large dataset of verbatim reports of sittings.

Integrations with other systems:

• Editor used to review the texts
• The chain of systems used to publish the verbatim report

Success metrics:

• Reduction of the time needed to produce the segments of the verbatim report
• Reduction of the time needed to produce the whole verbatim report