Career

Key Responsibilities

• Research, design, and develop Automatic Speech Recognition (ASR) systems for real-time, robust human-computer communication.

• Train, fine-tune, and optimize ASR models (e.g., DeepSpeech, Wav2Vec2, Whisper, Conformer, RNN-T, LAS) for multilingual and noisy environments.

• Build pipelines for speaker diarization, speech segmentation, and keyword spotting.

• Collaborate with TTS and LLM teams to create end-to-end conversational AI systems.

• Develop domain adaptation techniques (e.g., transfer learning, custom language models, acoustic modeling).

• Implement evaluation metrics (WER, CER, latency, robustness under noise).

• Research model efficiency and deployment techniques for low-latency ASR on cloud and edge devices.

• Integrate ASR into multimodal AI frameworks, combining speech, vision, and language understanding.
  
  

Requirements

• Bachelor’s, Master’s, or PhD in Computer Science, AI, Speech Processing, or related field.

• Strong programming skills in Python (C++ is a plus), with experience using PyTorch/TensorFlow.

• Solid knowledge of speech recognition, signal processing, and deep learning models.

• Hands-on experience with ASR frameworks (e.g., ESPnet, NVIDIA NeMo, Kaldi, Hugging Face Transformers).

• Understanding of acoustic modeling, language modeling, and decoding algorithms (CTC, WFST, attention-based decoding).

• Familiarity with multimodal learning and integration with LLMs/TTS.

• Knowledge of deployment optimization: quantization, streaming inference, GPU/TPU acceleration.

Preferred Qualifications

• Experience deploying real-time ASR systems in production.

• Background in multilingual ASR, code-switching, or low-resource languages.

• Research or practical experience with noise-robust and far-field ASR.

• Knowledge of self-supervised pretraining for speech (e.g., HuBERT, WavLM).

• Expertise in distributed training and large-scale speech model serving.