Highlights
Benchmark voice datasets, publish research findings, identify model gaps, work on Indic languages.
Description
Job Summary
pWe are seeking a Machine Learning Researcher to rigorously evaluate voice datasets, identify performance gaps across languages, and publish structured research findings. This role is ideal for someone passionate about speech-to-text and speech-to-speech models.
Responsibilities
- Benchmark voice datasets across ASR and speech models (Whisper, Deepgram, Google STT, Azure Speech).
- Measure performance using WER, CER, MOS, robustness, latency, and error pattern analysis.
- Design structured experiments to understand how dataset characteristics impact model accuracy.
- Systematically identify where speech models underperform across Indic languages and code-switching scenarios.
- Publish benchmarking findings as accessible blog posts and LinkedIn articles.
Required Skills
- Python
- Whisper toolkits
- WER, CER metrics
- Linguistic diversity analysis
- ML experimentation workflows
Required Skills Explained
- Python programming with a strong focus on ML experimentation workflows.
- Exposure to ASR and TTS systems, understanding model behavior, and running experiments.
- Familiarity with tools like PyTorch, TensorFlow, Whisper, SpeechBrain, Kaldi, etc.
- Experience in evaluating models using metrics such as WER, CER, MOS, SNR, etc.
- A deep interest and understanding of linguistic diversity, especially Indic languages.
Who is this for
pThis role is perfect for a curious researcher with hands-on experience in speech AI and a passion for linguistic diversity. You should have a strong analytical mindset and be comfortable with open-source tools.
Why This Job is a Good Opportunity
ulliWork on cutting-edge AI technologies in speech-to-text, speech-to-speech, and multimodal systems.liContribute to the growth of KGeN, an innovative company with substantial user base and revenue.liPublish your research findings and contribute to the broader AI community's knowledge base.liInteract with global clans and diverse datasets from more than 60 countries, enhancing your understanding of speech variations worldwide.liCreate a structured evaluation framework that will be pivotal in advancing AI technology for underrepresented languages.
Interview Preparation Tips
- Prepare examples of your previous research projects and publications, especially those related to speech models and linguistic diversity.
- Be ready to discuss your experience with Python, PyTorch, TensorFlow, Whisper, and other relevant ML tools.
- Think about real-world applications of voice datasets and how different accents or emotions can affect model performance.
- Discuss any work you have done on benchmarking models across multilingual data and identifying performance gaps.
- Explain your approach to evaluating audio clarity, speaker diversity, annotation accuracy, emotion depth, and accent/dialect coverage in datasets.
Career Growth in This Role
pIn this role, you will not only evaluate existing speech models but also contribute significantly to the development of new ones. Your insights into model performance gaps can directly influence future data collection strategies and even new research directions. As you grow in your career, you might move towards leading a team of researchers or take on more managerial responsibilities related to AI data strategy.pThe skills gained from this position will also open doors to roles in academia, where you can further explore the nuances of speech models across different languages and dialects. You may also find opportunities in tech companies that prioritize diversity in their datasets, ensuring fair representation for all communities.
Explore More Opportunities
Skills
Frequently Asked Questions
What is the ideal experience for this role?1-3 years of experience in speech AI or applied AI research.
What technical skills are required?Python, Whisper toolkits, WER and CER metrics, ML experimentation workflows.
Can I work remotely?Yes, remote options are available with regular team meetings.
