About the role
About the Opportunity
A fast-scaling AI voice technology firm operating in the speech AI and conversational AI space, we build and validate high-accuracy Automatic Speech Recognition (ASR) models for global enterprises. Our team ensures transcription quality for multilingual, real-world audio across domains—from customer service to media—using rigorous QA frameworks and proprietary tools to drive model improvement.
Role & Responsibilities
- Validate ASR output against ground-truth transcripts across multiple Indian languages and accents.
- Identify and categorize transcription errors (word substitution, omission, punctuation, timing) using standardized QA rubrics.
- Flag non-compliant, offensive, or biased content in transcribed audio for model sanitization.
- Test model performance across diverse audio environments—background noise, overlapping speech, low fidelity.
- Collaborate with ASR engineers to provide feedback that drives model retraining and accuracy uplift.
- Document QA metrics, trend insights, and edge-case patterns for process improvement.
Skills & Qualifications
- Must-Have
- ASR QA
- Transcription Validation
- Audio Quality Assessment
- Multilingual Linguistics (Indian languages)
- Speech Error Categorization
- QA Rubric Compliance
- Native-level fluency in at least 2 Indian languages
- Microsoft Excel
- Preferred
- Experience with speech datasets (Common Voice, LibriSpeech, etc.)
- Familiarity with NLP annotation tools (Prodigy, Label Studio)
- Basic understanding of speech-to-text model architectures
Benefits & Culture Highlights
- 100% remote work with flexible hours aligned to IST.
- Impactful role in shaping next-gen voice AI for global markets.
- Short-term contract with potential for extension based on performance.