Text & Speech Analyst

About the role

The Text & Speech Analyst role at this not-for-profit organization is essential for designing, developing, and refining automated speech recognition models and contextual text mapping techniques. This position also involves working closely with cross-functional teams to ensure that the models align with business requirements and objectives.

Responsibilities

• Design, develop, and refine automated speech recognition models and contextual text mapping techniques to ensure high-quality outputs for specific use cases.
• Employ techniques to guide and enhance Automated Speech Response and text harvesting models, ensuring that the AI interactions are effective and efficient.
• Develop effective AI text and speech services through proficient programming and utilization or ASR engines using standard and fit-for-purpose frameworks.
• Collaborate with cross-functional teams to ensure that the models are aligned with the business requirements and objectives.
• Conduct testing of variants and analyzing model behaviors, including deviations and degradations within given frameworks.
• Stay up to date with advancements in speech recognition and text harvesting technology and engineering best practices.
• Collaborate with product managers, data scientists, ML engineers, and business stakeholders to integrate AI models into business processes, products, or workflows.
• Create documentation and reusable libraries for internal and external use.
• Partner with the MLOps Engineer and other stakeholders to establish and implement observability and monitoring frameworks to adequately and timely identify degradations and potential ethical/bias issues.
• Establish and implement explainability frameworks.
• Provide recommendations for improvement in the areas of AI acceptable use and ethics, collaborating with Legal and Compliance to ensure adherence.
• AI Governance and Security: Ensure data quality and integrity as they apply to text and speech models. Ensure that data security protocols are followed in the definition and implementation of AI solutions in accordance with regulatory requirements and company policies. Develop safety filters and guide the ethical use of prompts to prevent biased or harmful responses.

Qualifications

• Working knowledge of ASR engines using frameworks like Wav2vec or Deep Speech.
• Working knowledge of Azure AI Services.
• Advanced programming knowledge, including mastery of programming languages such as Python, C++, Java, and especially AI-centric libraries like TensorFlow, PyTorch, and Keras.
• Cloud computing and knowledge for deploying and managing AI applications on cloud platforms like AWS, Google Cloud, or Microsoft Azure.
• Expertise in classical speech processing methodologies like hidden Markov models (HMMs), Gaussian mixture models (GMMs), Artificial neural networks (ANNs), Language modeling, etc.
• Functional knowledge of Snowflake and/or Databricks for AI capabilities and data management.
• Hands on experience on current deep learning (DL) techniques like Convolutional neural networks (CNNs), connectionist temporal classification (CTC) used for speech processing.
• Working knowledge of techniques for text parsing, sentiment analysis, and the use of transformers like GPT (generative pre-trained transformer) models.
• Exposure to open-source models such as Kaldi or OpenAI Whisper.
• Data management knowledge, including data pre-processing, augmentation, and generation of synthetic data, including the cleaning, labeling, and augmenting of data to train and improve AI models.
• Proficiency in Transformer Models like BERT-base. ELMo, ULM-FIT.
• Strong understanding of natural language processing concepts and techniques.
• Working knowledge of cloud services (i.e., MS Azure, AWS, Google Cloud).
• Experience with AI tools, such as MS Azure ML, Databricks AI, Snowflake CortexAI, Dataiku.
• Strong knowledge of data governance, data security, and compliance practices.
• Familiarity with data visualization and reporting tools (e.g., Webfocus, Power BI).
• Hands on experience with MLOps processes.