New Zealand English Media Audio Dataset

Home » Case Study » New Zealand English Media Audio Dataset

Project Overview:

Objective

Our mission is to curate a comprehensive dataset, the “New Zealand English Media Audio Dataset,” that empowers AI models to understand and interact with New Zealand English audio content effectively. This dataset will play a pivotal role in enhancing speech recognition, sentiment analysis, and language understanding for the New Zealand English dialect, making it a valuable resource for AI research and applications.

Scope

The project involves the collection and annotation of audio data from various sources to create a robust dataset. We aim to cover a wide range of topics, contexts, and accents within the New Zealand English language, ensuring that our dataset captures the nuances and diversity of this dialect.

Sources

Media Archives: We access a variety of media archives, including radio broadcasts, podcasts, news segments, and interviews, to gather authentic New Zealand English audio content.
Online Platforms: We collect audio content from online platforms, such as streaming services, YouTube channels, and social media, to ensure a comprehensive representation of spoken language in digital media.
Interviews: We conduct interviews with native New Zealand English speakers to create original audio content that reflects real-life conversations and interactions.

Data Collection Metrics

Total Audio Clips: Our dataset consists of a substantial volume of New Zealand English audio clips, providing a rich and diverse collection of spoken content.
Annotation Process: To enhance the dataset’s usability, each audio clip is annotated with valuable information, including sentiment, context, and speaker characteristics. This annotation process enriches the dataset and makes it a valuable resource for various AI applications.

Annotation Process

Stages

Total Audio Clips with Annotations: audio clips
Sentiment Labels: Positive, Neutral, Negative
Contextual Labels: News, Conversational, Educational, Entertainment
Speaker Characteristics: Age, Gender, Accent

Annotation Metrics

Total Audio Clips Annotated: Number of clips annotated with detailed linguistic features.
Sentiment Analysis: Annotations divided into Positive, Neutral, Negative sentiments.
Contextual Relevance: Clips annotated for context such as News, Conversational, Educational, Entertainment.
Speaker Details: Annotations include age, gender, and specific New Zealand accent characteristics.

Quality Assurance

Stages

Transcription Accuracy Check: Rigorous review process for verifying the accuracy of transcriptions.
Metadata Validation: Ensure each audio sample is correctly tagged with relevant metadata.

QA Metrics

Transcription Accuracy: Number of audio clips checked and corrected for transcription accuracy.
Metadata Consistency: Verification count of audio clips with accurately tagged metadata including speaker details and context.
Sample Validation: Number of samples validated for correct sentiment annotation.

Conclusion

The “New Zealand English Media Audio Dataset” is a pivotal resource for researchers, developers, and AI enthusiasts interested in advancing the understanding of the New Zealand English dialect. With its extensive collection of annotated audio clips and meticulous metadata, this dataset empowers the development of cutting-edge speech recognition, sentiment analysis, and language understanding models. It serves as a catalyst for breakthroughs in AI applications tailored to the New Zealand English context, fostering innovation, and bridging language barriers.