Data Engineer (Speech & LLM)

As a Data Engineer, you will own the end-to-end speech data lifecycle that powers advanced speech and speech-to-speech models. Your primary responsibility is to build, curate, validate, and deliver high-quality datasets that enable robust speech understanding, generation, and conversational interaction.

This role is critical to ensuring that models are trained on clean, diverse, well-annotated, and model-ready data.

Key Responsibilities

Speech Data Curation

· Build datasets supporting:

o Speech recognition and understanding

o Multilingual and code-mixed speech

o Conversational and dialog-style speech

o Speech generation and synthetic voice data

o Audio-to-audio conversational scenarios

· Prepare datasets that capture:

o Speaker variation and continuity

o Emotional and expressive speech cues

o Real-world noise and acoustic conditions

o Conversational turn structure and timing

Ensemble-Based Data Curation

· Implement data curator pipelines using outputs from:

o Multiple in-house speech models

o External or open-source speech models

· Aggregate, reconcile, and validate model outputs to:

o Generate reliable annotations

o Filter low-confidence samples

o Detect inconsistencies and label noise

· Apply rule-based and confidence-driven selection strategies.

Validation & Quality Control

· Perform automated validation for:

o Audio integrity and format consistency

o Transcript alignment and correctness

o Language and speaker metadata accuracy

· Run sampling-based manual audits.

· Produce dataset quality reports and summaries.

Engineering & Operations

· Build scalable, reproducible data pipelines in Python and C++.

· Handle large audio corpora efficiently on Linux systems.

· Generate training-ready manifests and metadata.

· Maintain dataset versions, lineage, and reproducibility.

Required Skills

· 3+ years of experience in data engineering or ML data pipelines.

· Strong Python skills for large-scale data processing.

· Experience working with audio or speech datasets.

· Familiarity with annotation formats and metadata schemas.

· Knowledge of Linux, Bash, and Git workflows.