Role Overview
VoiceArena builds benchmarks, datasets, and AI systems that enable machines to communicate naturally, powering some of the world's most widely used speech models.
Responsibilities
- Design and manage end-to-end data delivery workflows from raw ingestion to structured output.
- Ensure reliable, versioned, and reproducible data handoffs to ML training infrastructure.
- Monitor pipeline health, resolve bottlenecks, and optimize throughput for large‑scale data movement.
- Design and run evaluations for ASR and speech-to-speech systems.
- Benchmark leading speech models to identify real‑world strengths, weaknesses, and failure modes.
- Build evaluation frameworks that guide top AI labs on model performance and improvement areas.
- Fine‑tune speech recognition models (e.g., Whisper, wav2vec2, HuBERT) targeting Word Error Rates of ~5%.
- Experiment with multilingual, code‑switched, accented, and noisy speech data.
- Work with large‑scale, production‑grade speech datasets across 20+ Indian languages using PyTorch, TensorFlow, or Hugging Face Transformers.
Requirements
- Students in B.Tech/B.E. graduating in 2027 (CSE, EE, AI/ML or related fields)
- Strong interest in speech, audio, NLP, or multimodal AI
- Hands‑on experience fine‑tuning speech or language models (Whisper, wav2vec2, HuBERT, etc.)
- Hands‑on experience building speech‑driven pipelines, classifiers, or assistants
- Hands‑on experience working with PyTorch, TensorFlow, or Hugging Face Transformers
- Hands‑on experience using cloud platforms (AWS, GCP, Azure) and container tools (Docker, Kubernetes)