Speech Data Project Manager

At 42dot, we're building performance evaluation systems for in-service speech recognition and developing comprehensive training and evaluation datasets for our LLM modules through meticulous data annotation. We strategically collect TTS voice data to ensure a diverse range of authentic, high-quality audio samples. Additionally, we are at the forefront of defining our philosophy on voice design in automotive environments, integrating robust acoustic and user experience principles tailored specifically for vehicle settings. This role will participate in dataset collection and validation to mitigate issues such as data bias and errors.

Responsibilities

• Verification of Speech Data

  • Validate speech data related to STT, TTS and wake-up word detection to ensure accuracy and consistency.

• TTS Data Collection Strategy & Execution

  • Design and implement data collection strategies that reflect North American linguistic and cultural characteristics.

  • Secure high-quality English, Spanish, and French text and speech data from diverse sources (e.g., online media, audio archives, user interviews).

• Data Quality Control

  • Review collected data for pronunciation, intonation, grammar, and vocabulary accuracy to ensure suitability for model training.

  • Perform outlier detection and data cleaning tasks (e.g., noise removal, audio clipping, text normalization).

• Process Automation & Optimization

  • Develop scripts and tools (using Python, R, etc.) to automate repetitive tasks in data collection and verification.

  • Build and manage data pipelines and propose workflow improvements to optimize the process.

• Outsourcing Management

  • Oversee and manage outsourcing agencies responsible for speech data labeling, ensuring adherence to quality standards and deadlines.

• Collaboration & Communication

  • Work closely with development teams, speech engineers, and language experts to set data quality standards and project objectives.

  • Provide regular reports on project progress, challenges, and improvement measures.

• Market & User Analysis

  • Analyze language usage trends, dialects, and intonation patterns in North America to continuously refine data collection strategies.

  • Incorporate user feedback and emerging research trends to update and improve the datasets.

Qualifications

• Experience

  • Over 3 years (or equivalent experience) in voice signal-related roles, including speech data verification, labeling, and managing outsourcing agencies.

  • Proven experience in collecting and validating speech data for various audio signal tasks.

• Educational Background

  • A Master’s or Doctoral degree in Linguistics, Speech Signal Processing, Computer Science, Data Science, or a related field.

• Language & Communication Skills

  • Native-level proficiency in at least two of the following languages—English, French, and Spanish—is a plus.

  • Strong understanding of North American dialects and cultural nuances is required.

  • Excellent documentation, presentation, and teamwork skills.

  • Professional-level Korean language proficiency is an asset for research collaboration.

• Project Management & Problem-Solving

  • Strong analytical, problem-solving, and project management skills, with the ability to handle multiple tasks and set priorities effectively.

Preferred Qualifications

• Specialized Industry Experience

  • Proven track record in quality control and management of audio data labeling projects.

  • Strong understanding of technologies related to STT, TTS and wake-up word detection.

  • Project experience in this field and hands-on experience with deep learning frameworks such as TensorFlow and PyTorch.

• Data Management Expertise

  • Experience in building and managing large-scale multi-modal (text + speech) datasets and optimizing data cleaning processes.

• Sound Engineering & Narration Directing Expertise

  • Demonstrated experience in sound engineering, including designing and optimizing acoustic environments, implementing advanced audio processing techniques,