Audio AI Engineer

ZVC SINGAPORE PTE. LTD.; Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

Zoom is seeking an Audio AI Engineer to research and develop low-latency audio algorithms for accent conversion, voice conversion, TTS, and ASR, shaping real-time streaming experiences.

You will prototype end-to-end models, optimize for quality and latency, and collaborate with product and platform teams to deploy in real-time communication systems. A PhD or equivalent experience is preferred.

Qualifications

• PhD or equivalent in a relevant field such as streaming, accent or voice conversion, TTS or ASR.
• Experience with low-latency, real-time audio systems and streaming architectures.
• Strong proficiency in PyTorch or TensorFlow and Python/C/C++.

Responsibilities

• Research, design, and develop algorithms for accent and voice conversion, TTS, and ASR in low-latency systems.
• Prototype end-to-end audio models to improve intelligibility and naturalness while preserving speaker identity.
• Collaborate with product/platform teams to deploy models into real-time video/audio communications.
• Analyze and optimize model performance for quality, latency, robustness, and scalability.
• Stay current with speech-processing advances and contribute to patents and internal knowledge sharing.

Skills

Python

C/C++

PyTorch

TensorFlow

Transformers

Streaming architectures

Model compression

Education

PhD or equivalent in Streaming, Accent Conversion, Voice Conversion, TTS, or ASR

Tools

PyTorch

TensorFlow

Python

C/C++

Job description

What you can expect

As an Audio AI Engineer, you will research and develop algorithms for accent conversion, voice conversion, speech synthesis, and speech recognition on low-latency streaming architectures. You'll prototype and refine end-to-end audio models that enhance intelligibility and naturalness while maintaining speaker identity. Working closely with product and platform teams, you'll help bring these models into real-time communication systems. You will also evaluate and optimize model performance across dimensions such as quality, latency, and scalability. Staying current with advances in speech processing, you'll contribute to innovation through patents and internal knowledge sharing.

About the Team

Zoom's Audio team develops real-time audio features based on AI algorithms. Members of the team are spread worldwide, including the U.S., China and Singapore.

Responsibilities

• Researching, designing, and developing algorithms for accent conversion, voice conversion, speech synthesis, and automatic speech recognition, focusing on low-latency streaming architectures.
• Prototyping end-to-end audio models that enhance intelligibility and naturalness while preserving speaker identity and expressiveness.
• Collaborating closely with product and platform teams to integrate models into real-time video and audio communication systems.
• Analyzing and optimizing model performance across speech quality, latency, robustness, and scalability dimensions.
• Staying current with the latest developments in speech processing research, and contribute to the community through patents, and internal knowledge sharing.

What we're looking for

• Hold a PhD or equivalent experience in a relevant field in Streaming, Accent Conversion, Voice Conversion, TTS, or ASR. More than 2 years of relevant industry experience considered a plus.
• Show proficiency in deep learning frameworks like PyTorch or TensorFlow.
• Demonstrate effective programming skills in Python, C/C++, or similar languages.
• Have an understanding of sequence modeling architectures (Transformers, RNNs, diffusion models, or conformers).
• Demonstrate experience developing and deploying low-latency, real-time speech or audio models with streaming architectures and optimized pipelines.
• Show familiarity with model compression and acceleration techniques, including quantization, pruning, and distillation.
• Exhibit experience working with real-time audio systems in networked communication environments.
• Publish in top-tier conferences such as Icassp, Interspeech, NeurIPS, and ICLR.