CaptureLabz Nahuatl & Mixtec Pilot

CaptureLabz Research Program

CaptureLabz is being developed as an applied research program within the CaptureLabz voice dataset initiative. The program investigates how structured speech datasets affect the reliability, robustness, and governance of modern speech recognition and language technology systems. Its work focuses on dataset methodology, acoustic-condition variation, and the development of speech resources for languages that remain underrepresented in current AI training pipelines.

• CaptureLabz Overview
• Recording Protocol
• Speech Robustness Benchmark
• Nahuatl & Mixtec Pilot Study

Primary Research Objectives

• Evaluate how structured multi-condition speech datasets influence ASR robustness
• Develop reproducible dataset capture methods for low-resource languages
• Document governance practices for ethically sourced voice data
• Create dataset structures suitable for AI training, benchmarking, and language preservation

Current Research Components

• Recording Methodology — controlled multi-condition speech capture
• Robustness Benchmark — experiments evaluating model performance under acoustic variation
• Language Dataset Pilots — initial structured datasets for Nahuatl and Mixtec

These components are documented through the CaptureLabz research documentation within XPGuess Learn and are intended to support both academic research and applied speech technology development.

What This Pilot Is

The CaptureLabz Nahuatl and Mixtec pilot is an early validation dataset designed to test whether structured speech collection can produce more useful training and evaluation material for low-resource language research. The pilot is not meant to be a final corpus for every dialect or use case. It is meant to establish a disciplined starting point.

That starting point includes controlled recording conditions, standardized metadata, condition-aware dataset packaging, and a research design that allows later model comparison rather than simple archiving alone.

Why Nahuatl and Mixtec

Nahuatl and Mixtec matter both culturally and technically. They are spoken by large populations relative to many endangered languages, yet they remain significantly underrepresented in modern speech technology systems. That makes them strong pilot languages for testing whether a more structured approach to speech data can reduce language exclusion in AI systems.

The pilot is also grounded in real community access. That matters because low-resource dataset design is often blocked not by theory, but by lack of trusted speaker participation and lack of structured collection workflows.

Pilot Structure

The pilot is designed as a dataset-validation phase rather than as a finished language archive. Its purpose is to test the CaptureLabz collection workflow, file structure, and metadata logic using a practical speech set that can later be expanded.

Even if the pilot begins with modest scale, the structure matters more than raw size at this stage. Researchers need to know how the data was created, not just how much exists.

Recording Conditions

The pilot uses the CaptureLabz four-condition recording protocol so that the same word or phrase does not exist in only one acoustic state.

This structure allows later comparison between conventional narrow-audio training and multi-condition training. It also creates a more realistic base for testing speech model behavior under acoustic distribution shift.

Metadata and Governance

The value of the pilot depends on more than audio files. It also depends on metadata, contributor awareness, and collection structure. CaptureLabz is designed to package speech with enough context for later research use rather than leaving recordings disconnected from provenance.

Example metadata fields

speaker_id
language
dialect
recording_condition
environment
device_type
session_id
transcript
ipa
consent_status
qa_status

Within the broader XPGuess / NTG framing, the pilot also aims to preserve collection logic. That means documenting who recorded, under what conditions, and with what structure. This is essential if the dataset is later used for benchmark comparison or model analysis.

Research Questions

The pilot supports several early-stage research questions:

• Can a small multi-condition dataset improve speech-model robustness compared with narrow clean-audio training sets?
• Which recording conditions contribute most to performance stability under acoustic shift?
• How important is metadata and condition labeling for reproducible low-resource language research?
• Can structured pilots serve as a practical bridge between community participation and research-grade dataset design?

These questions are intentionally practical. The pilot is meant to provide a controlled foundation that can be expanded later rather than trying to answer every linguistic or technical question in one release.

Why This Pilot Matters

Many language projects stop at documentation. Many speech projects stop at raw collection. The CaptureLabz pilot tries to move one step further by testing whether underrepresented-language speech can be collected in a form that is immediately more useful for research, benchmarking, and later model evaluation.

That makes the Nahuatl and Mixtec pilot important not only because of the languages themselves, but because of what the pilot is trying to prove: that low-resource language data can be collected with enough structure to matter technically, not just symbolically.

Terminology, Frameworks, and Foundational Work

XPGuess — Extended Performance Guessing — is an educational decision-learning construct used to explore how development paths and outcomes unfold over time.

Natural Technical Governance (NTG) documents training and participation using first principles rather than subjective opinion.

The conceptual foundations derive from earlier technical work by Michael A. Piña, including biomechanical and developmental research.

Reference: “Beginning and Staying with the Basics: Building from the Ground Up”

Additional work: Coach Teaches Animals: Gymnastics Stretching