Voice AI / Speech ML Engineer

Job Overview

• We are looking for a strong Voice AI / Speech ML Engineer who can work on fine-tuning and improving speech models for real-world users across different accents, dialects, noisy audio environments, and multilingual conversations.
• This role is focused on making speech AI systems work beyond clean demo conditions. The engineer will work on speech-to-text, text-to-speech, language model layers where required, and audio data pipelines for low-resource or multi-dialect use cases.
• The ideal candidate should have hands-on applied ML experience, strong exposure to speech models, and practical experience with messy audio data, model fine-tuning, evaluation, and production deployment.

Key Responsibilities

• Fine-tune speech-to-text, text-to-speech, and related voice AI models for real-world accents, dialects, and noisy audio conditions.
• Work on audio data sourcing, cleaning, labeling, augmentation, and preparation for model training.
• Build and improve training pipelines for low-resource, multilingual, and multi-dialect speech use cases.
• Evaluate model performance using practical metrics such as word error rate, naturalness, robustness to noise, and code-switching accuracy.
• Test models on real-world audio instead of relying only on standard benchmarks.
• Improve model performance based on production feedback and user audio samples.
• Work with modern speech model architectures such as Whisper-family models, neural TTS, ASR/STT models, and speech-to-speech systems.
• Optimize models for latency, accuracy, and real-time serving where required.
• Collaborate closely with engineering and product teams to ship voice AI features quickly.
• Own the model lifecycle from data preparation and training to evaluation and production improvement.

Ideal Candidate Profile

• 3+ years of experience in applied ML, speech AI, voice AI, ASR/STT, TTS, or related ML engineering roles.
• Hands-on experience fine-tuning and deploying ML models in production.
• Strong experience with Python, PyTorch, and ML training workflows.
• Experience working with ASR/STT, TTS, speech recognition, speaker/audio processing, or voice AI systems.
• Practical experience with audio data collection, cleaning, labeling, augmentation, and preprocessing.
• Strong understanding of model evaluation, especially for real-world speech performance.
• Experience measuring WER, latency, robustness, and model quality across different audio conditions.
• Comfortable working with noisy, imperfect, and low-resource datasets.
• Ability to own technical work independently and ship fast.
• Strong problem-solving ability and practical engineering mindset.

Good to Have

• Experience with low-resource languages, multilingual speech, or multi-dialect speech systems.
• Experience with Whisper, wav2vec, Conformer, DeepSpeech, FastSpeech, Tacotron, VITS, or similar speech architectures.
• Experience with code-switching, accent adaptation, or dialect-specific model improvement.
• Experience deploying speech models for real-time or near-real-time use cases.
• Exposure to MLOps, model monitoring, inference optimization, or production ML pipelines.
• Interest in voice AI, emerging markets, and real-world speech technology.

You Are a Good Fit If

• You have fine-tuned speech or ML models and shipped them to production.
• You understand that real-world audio data is messy and requires strong preprocessing.
• You can evaluate models using practical business and product metrics, not only benchmark scores.
• You are comfortable working independently with imperfect data.
• You can improve model performance through data, training, evaluation, and production feedback loops.

You May Not Be a Fit If

• You have only used speech APIs without training or fine-tuning models.
• You have not worked with audio data pipelines or model evaluation.
• You are looking for a research-only role without production responsibility.
• You are not comfortable working with noisy, incomplete, or low-resource datasets.
• You have not handled applied ML systems end to end.