Speech Processing ML Algorithm Engineer

At Apple, new ideas have a way of quickly becoming extraordinary products, services and customer experiences. Sound is an essential and compelling facet of the customer experience - Apple has assembled a world-class Acoustics team that enables our customers to experience music with delight, communicate with clarity and appreciate our products without disturbance from noise. In Apple Acoustics we work with obsessive attention to detail, directly contributing to products that ship to millions of people around the world.

The Acoustic ML Algorithm Development team sits in Apple's Hardware organization and develops the new architectures and training paradigms that define the audio experience of the next generation of Apple hardware along with our counterparts in the Software organization. We also work closely with various teams behind microphones, loudspeakers, wireless communication, silicon, etc and we focus on features that improve the customer experience across speech, music, and general audio. We are searching for a Speech Processing ML Algorithm Engineer to advance our work in the domain of speech enhancement, with an emphasis on low-latency processing applications. As a Speech Processing ML Algorithm Engineer on the Acoustics ML Algorithm Development team, you will design, train, and evaluate models for speech enhancement. Examples include noise suppression, de-reverberation, echo suppression, and general multi-microphone processing under the low-latency, real-time constraints of shipping hardware. You will work with cross functional teams to bring early prototypes through to production.MS or PhD in a computational science field or 3+ years of experience in the field of ML for speech processing. A passion for audio ML research for applied product applications, with a deep understanding of transformers, recurrent and convolutional neural networks, etc. Experience designing, training, and evaluating machine learning models for speech enhancement, such as noise suppression, dereverberation, source separation, or echo residual suppression. Experience building models that meet low-latency, real-time requirements, including streaming and causal processing, and a clear understanding of the quality, complexity, and latency trade-offs involved. Strong grounding in audio and speech signal processing fundamentals, for example STFT analysis/synthesis. Proficiency with PyTorch and Bash, including version control, code review, testing, and reproducible experiments. A habit of following the state of the art literature closely, with the ability to reproduce, critique, and build on published results. Working knowledge of speech quality evaluation, spanning objective metrics (e.g. PESQ, STOI, SI-SDR, DNSMOS) and subjective listening tests, along with the data simulation and augmentation needed to support them.Experience applying machine learning to adaptive filter prediction and control, such as echo cancellation, active noise control, or adaptive beamforming, including hybrid classical and learned systems. Familiarity with Lightning and Hydra. Familiarity with model efficiency techniques such as quantization-aware training or distillation. Experience in high-performance cloud computing for model training. Open source contributions to a repository using in the ML audio community.