QA Specialist for Audio Annotation and Diarization (Russian)

Role Overview:

This role focuses on building a highly accurate, evaluation-grade dataset of transcribed, multi-channel audio recordings to assess multilingual, multi-speaker AI systems. You will evaluate high-quality conversations that represent diverse dynamics, contexts, and demographics, ensuring the integrity and quality of both audio recordings and their human-verified annotations.

Key Responsibilities

Audio Quality Assurance:

• Evaluate multi-channel audio recordings to ensure they meet strict technical and fidelity requirements.
• Verify channel isolation to ensure no audio bleed and confirm that recordings were captured in appropriate, quiet environments free from disruptive background noise, clipping, or low gain.

Transcription & Diarization Verification:

• Review human-validated transcriptions to guarantee exceptionally high accuracy and adherence to strict low error-rate (WER) targets.
• Confirm that transcripts accurately capture spontaneous, unnormalized speech, preserving natural conversational dynamics such as overlaps, interruptions, and false starts.
• Validate the precision of turn-level and word-level timestamps, as well as speaker identification, paying special attention to complex, overlapping dialogue, while comfortably reading and validating the underlying JSON-formatted data to ensure accurate metadata tagging and timestamp logic.

Metadata & Content Review:

• Verify the accuracy of all applied metadata, including demographic markers, contextual domains, and specific conversational tags.
• Enforce strict safety and privacy standards by auditing sessions to ensure no Personally Identifying Information (PII), toxic, or sensitive content is present.

Execution & Reporting:

• Assess the end-to-end quality of the annotation task, assigning clear pass/fail or agree/disagree statuses during your review.
• Provide detailed, actionable comments and feedback whenever disagreeing with an annotator''s work.

Requirements & Qualifications

• Exceptional ear for audio fidelity and the ability to detect subtle background noises, channel bleed, or clipping.
• Meticulous attention to detail for verifying word-level timestamps and strict, unnormalized verbatim transcription rules.
• Native proficiency in Russian.
• Ability to accurately assess complex multi-speaker dynamics.

Ideal Backgrounds include:

• Linguists/Phonetics Experts: Deep understanding of natural, unnormalized speech patterns. Expertise in accurately identifying and annotating complex conversational dynamics, including overlaps, false starts, and backchannels.