We are looking for an experienced speech scientist to join our team at Rev. You must be comfortable and well-versed with building current ASR and NLP solutions and up to date with the latest developments within industry and academia. You enjoy working with the latest machine/deep learning technologies and implementing the latest research findings.
Responsibilities:
As a Senior Speech Scientist, you will
- work with a team of engineers and researchers to improve and innovate on the existing ASR and NLP infrastructure.
- Help develop new (speech) translation and generative audio pipelines
- evaluate and benchmark existing ASR and NLP models.
- Experiment and discover creative solutions to difficult problems
- expand and prototype novel ASR and NLP solutions - improve word accuracy, distinguish and leverage speaker characteristics, and dynamically fine-tune and model speech in different acoustic environments.
- innovate new approaches and product features.
- automate and integrate workflow from diverse systems.
- interact daily with other teams at Rev working towards a shared goal
Qualifications:
- University degree in Computer Science, Software Engineering, or related fields
- 3+ years of experience supporting and working on production ML systems (training models and tuning existing systems)
- Fluency in Python, C++, shell scripting, and Linux usage.
- Broad mastery of ASR or NLP techniques such as neural net architectures (Transformer / LSTM / CTC / Transducer), acoustic and language models, and decoding.
- Experience with Deep Learning frameworks (such as TensorFlow or PyTorch) and training large models.
- Excellent oral and written communication skills.
- Comfortable working with remote teams as a proactive team member.
Nice to have knowledge of:
- Large Language Models (LLMs), especially training, finetuning, and inference
- Efficient training techniques
- Monitoring production model performance
- Different optimizers for model training
- Fine tuning and knowledge distillation
- Data forensic and conditioning
- Low-resource languages ASR techniques