ASR Engineers / Machine Learning Researchers

Company Description

Friska.AI redefines healthcare with advanced artificial intelligence, delivering tailored health solutions designed uniquely for individuals. Our mission is to empower people and communities with personalized healthcare recommendations that address specific needs and promote healthier lives. We analyze individual health data to create customized plans for nutrition, diet, fitness, and more. Detailed health reports provide actionable insights to achieve and sustain better health. Our platform also offers tools for population health management, aiding organizations and healthcare providers in understanding community health trends.

Role Description

This is a full-time remote role for ASR Engineers / Machine Learning Researchers. The role involves developing algorithms, working on pattern recognition and machine learning models including neural networks, and applying statistical analysis to improve our AI-driven health solutions. Day-to-day tasks include collaborating with cross-functional teams to design and implement AI models, analyzing large datasets, and continuously enhancing the performance and accuracy of our healthcare recommendations through innovative research.

Experience:

• Minimum 2 years of hands-on experience in ASR, speech processing, or closely related ML/NLP domains.
• Prior work with impaired speech datasets or accessibility-focused technologies is highly preferred.
• Background in projects involving assistive tech, voice accessibility, or inclusive AI will be considered a strong advantage.

Qualifications

Programming & Frameworks:

• Proficiency in Python.
• Hands-on experience with PyTorch and/or TensorFlow for model development.

ASR Expertise:

• Practical experience with advanced ASR models like Whisper, Conformer, Wav2Vec 2.0, etc.
• Familiarity with speaker-adaptive and impaired speech models is a strong plus.

Model Optimization:

• Knowledge of optimization methods such as quantization, pruning, distillation, and LoRA (Low-Rank Adaptation) for lightweight deployment.

Data Handling:

• Proficiency in handling large-scale speech datasets.
• Strong understanding of data preprocessing, augmentation, and labeling pipelines for robust model training.

Evaluation:

• Experience with ASR evaluation metrics: WER (Word Error Rate), CER (Character Error Rate), RTF (Real-Time Factor).
• Knowledge of SER (Sentence Error Rate) and PER (Phoneme Error Rate) is desirable.

Engineering Practices:

• Familiarity with software engineering workflows (e.g., Git, GitHub/GitLab, CI/CD pipelines).
• Ability to build and integrate RESTful APIs for deploying ASR functionalities.

Cloud & Deployment:

• Experience with cloud platforms such as AWS, Google Cloud, or Azure for training, hosting, and scaling models.

Responsibilities:

• Design, train, and optimize speaker-adaptive ASR models, focusing on high accuracy and real-time performance, especially for users with speech impairments.
• Apply domain adaptation techniques (e.g., fine-tuning, transfer learning, LoRA) to customize models across diverse speech patterns and demographics.
• Build feedback-based learning systems to continually refine ASR performance based on user corrections and interactions.
• Collaborate closely with frontend/backend developers, UI/UX designers, and clinical experts to deliver user-centered, accessible speech solutions.
• Rigorously test and benchmark models using both standard evaluation metrics and real-world test environments.
• Stay up-to-date with state-of-the-art research in ASR and bring innovative ideas into the development pipeline.