AI Trainer (Contractor), Voice and Speech Evaluation - USA

The role

You will listen to speech samples to judge naturalness, expressiveness, and faithfulness to instructions. You will score voices against rubrics and label voice attributes like gender, age, accent, and tone.

What you will do

• Listen to paired or grouped speech samples and judge naturalness, expressiveness, or faithfulness to instructions
• Score voices against rubrics for naturalness, clarity, emotion, pacing, and delivery style
• Label voice attributes such as perceived gender, age, accent or dialect, and tone
• Flag mismatches with existing metadata
• Judge how closely cloned voices match their source speaker
• Review batches of new voices for audio quality and suitability before public release
• Test new speech features in pre-release environments and report bugs, edge cases, and usability issues
• Write clear, structured findings that engineers can act on
• For multilingual roles: evaluate pronunciation, dialect accuracy, and naturalness in your native language

Required qualifications

• Background in audio production, broadcast, voice acting, music, linguistics, phonetics, or similar
• Ability to hear and describe subtle differences in prosody, timing, pronunciation, and audio quality
• Consistent, careful judgment across large batches of similar samples
• Comfort with beta web tools, VPN access, and spreadsheets
• Strong written English for reporting findings
• Ability to turn tasks around within a day when needed, including occasional weekend availability

Nice to have

• Native or near-native fluency in Hindi, Tamil, Telugu, Bengali, Marathi, Brazilian Portuguese, or a regional Spanish dialect
• Experience with TTS, voice AI, dubbing, or localization QA
• Familiarity with A/B evaluation or annotation work

About the company

Inworld is a research lab and inference provider focused on realtime AI for consumer-facing applications. They build first-party speech models, serve LLMs, and run inference for high-volume, realtime workloads.