Alexa Wake Words in Canadian French (Adults)

Home » Case Study » Alexa Wake Words in Canadian French (Adults)

Project Overview:

Objective

Our goal was to develop a comprehensive dataset that would significantly improve the performance of wake word detection systems and voice assistants designed for the Canadian French-speaking demographic. Specifically, this dataset, titled Alexa Wake Words in Canadian French (Adults), aims to enhance the accuracy and efficiency of voice recognition technologies tailored for adult users in this linguistic group. By focusing on this key demographic, we strive to ensure that voice assistants can better understand and respond to wake words in Canadian French, thereby providing a more seamless and effective user experience.

Scope

We undertook the collection of a wide array of audio recordings from native Canadian French speakers. Our diverse data set encompassed various environments and accents, all annotated with precise wake word markers.

Sources

Voice Assistant Users: We collaborated with Alexa users fluent in Canadian French, gathering audio clips of them using the wake word in different contexts.
Voice Actors: We employed professional voice actors proficient in Canadian French to generate synthetic wake word recordings, thereby enriching the dataset’s diversity.
Public Domain Recordings: We sourced publicly available audio containing the “Alexa” wake word in Canadian French.

Data Collection Metrics

Total Audio Clips Collected: 30,000
User Contributions: 15,000
Voice Actor Recordings: 10,000
Public Domain Extracts: 5,000
Total Audio Clips Annotated: 30,000

Annotation Process

Stages

Wake Word Annotation: We meticulously marked the temporal boundaries of the “Alexa” wake word in each audio clip.
Speaker Demographics: We gathered metadata about the contributors, including age, accent, and gender.
Recording Conditions: We documented the recording conditions such as ambient noise levels and the devices used.

Annotation Metrics

Audio Clips with Wake Word Annotations: 30,000
Speaker Demographics: 30,000
Recording Condition Metadata: 30,000

Quality Assurance

Stages

Annotation Verification: We implemented a rigorous validation process using both automated tools and human reviewers to ensure the accuracy of wake word annotations.
User Consent: We ensured that all user-contributed audio clips included explicit consent for usage in the dataset, with all personally identifiable information anonymized.
Privacy Compliance: We adhered strictly to privacy regulations, including data protection policies, and provided mechanisms for contributors to opt-out or request data removal.

QA Metrics

Annotation Validation Cases: 3,000 (10% of total)
Privacy Audits: 15,000 (for user-contributed data)

Conclusion

This dataset, the Alexa Wake Words Dataset in Canadian French (Adults), demonstrates our capability in collecting and annotating high-quality speech data. It stands as a valuable resource for enhancing voice recognition and natural language processing technologies, specifically catering to the Canadian French-speaking adult demographic. Our commitment to diversity, meticulous annotation, and stringent privacy compliance underscores our dedication to advancing machine learning research and development.