Sony Research India is seeking a dynamic and motivated Speech Recognition Intern to join our innovative research team. As an intern, you will work on real-world problems in automatic speech recognition (ASR), focusing on improving noise robustness and reducing hallucinations in transcription outputs. You'll gain hands-on experience with state-of-the-art tools and datasets, and contribute to impactful projects alongside experienced researchers and engineers.

Key Responsibilities:

• Explore and develop techniques to enhance ASR robustness under noisy, low-resource, and domain-shifted conditions.
• Investigate hallucination phenomena in end-to-end ASR models (e.g., Whisper, Wav2Vec2, etc.) and propose mitigation strategies.
• Conduct experiments using large-scale speech datasets and evaluate ASR performance across varying noise levels and linguistic diversity.
• Contribute to publications, technical reports, or open-source tools as outcomes of the research.

Work Location:

• Remote

Duration of the paid Internship:

• This paid internship will be for a period of 6 months starting June first week of 2025.
• 9:00 to 18:00 (Monday to Friday).

Qualification:

Currently pursuing/completed Masters in (Research) or Ph.D. in deep learning/machine learning with hands-on experience on Transformer models with an applications audio/speech.

Must Have Skills:

• Strong programming skills in Python, and familiarity with PyTorch or TensorFlow.
• Experience with speech processing libraries (e.g., Torchaudio, ESPnet, Hugging Face Transformers).
• Prior experience with ASR models like Wav2Vec2, Whisper, or RNN-T is a plus.
• Ability to read and implement academic papers.
• Strong foundation in machine learning and signal processing.

Good to have skills:

• Familiarity with prompt tuning, contrastive learning, or multi-modal architectures.
• Experience with evaluating hallucinations or generating synthetic speech/audio perturbations.