AI Accent Adaptation Fine-tuning Engineer

Nelvix Labs

Remote · $80k – $100k / year · Full-time · Posted 4 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

A speech recognition model trained primarily on American English performs substantially worse on Indian English, Nigerian English, and Scottish English — even though all four are grammatically correct and linguistically rich. Accent bias in ASR isn't a marginal issue; it's a systematic failure that affects hundreds of millions of speakers and makes voice products feel exclusionary to large user segments. We provide speech AI infrastructure to 80 enterprise clients across call centers, transcription platforms, and voice assistants. Accent adaptation is the top feature request from clients serving global user bases, and we're building the systematic capability to address it. You'll own the accent adaptation fine-tuning pipeline: curating and augmenting accent-specific training data, designing fine-tuning strategies that improve performance on underrepresented accents without degrading accuracy on the base distribution, and building the evaluation benchmarks that track accent-specific WER across our supported accent groups. The technical challenge is real: accent-specific data is scarce, and transfer learning strategies that work for domain adaptation don't always generalize to phonetic adaptation.

Fully remote.

Responsibilities

Build accent-specific fine-tuning pipelines for ASR models targeting underrepresented accent groups
Curate and augment accent-specific training data from available sources and synthetic generation
Design evaluation benchmarks measuring WER per accent group across production model versions
Implement data augmentation strategies to improve coverage for low-resource accent groups
Analyze error patterns to identify phonetic features driving degraded accuracy per accent
Validate fine-tuned models against holdout accent data before production deployment

Requirements

3+ years in speech processing, NLP, or audio ML engineering
Hands-on experience fine-tuning ASR or TTS models with HuggingFace or similar
Understanding of phonetics, accent variation, and low-resource learning challenges
Strong Python and PyTorch skills with audio data processing experience
Familiarity with WER evaluation and audio dataset management