Job Description:
• Define how Innodata designs, structures, and evaluates audio data for speech and audio models
• Translate speech and audio model requirements into data specifications covering modalities, transcription and annotation schemas, sampling, and evaluation criteria
• Build evaluation methodologies covering semantic accuracy, noise and accent robustness, code-switching, diarization error rate, speech naturalness and intelligibility, and streaming latency
• Structure, enrich, and sample audio across languages, accents, acoustic conditions, demographics, emotional and paralinguistic range, scripted and spontaneous speech, and speaker configurations
• Partner with transcription and linguistics leads to develop transcription specifications and quantify their impact on model results
• Partner with audio solutions and engineering teams to ensure collected audio meets model objectives
• Run fine-tuning and evaluation experiments with ablations linking data choices to measurable improvements
• Design adversarial and stumping evaluations to identify speech-system failures
• Publish benchmarks, methodology, and research papers
• Work with annotation teams, subject-matter experts, and synthetic- and augmented-audio pipelines to operationalize specifications
• Partner directly with customers and frontier labs building ASR, text-to-speech, speech-to-speech, conversational voice, diarization, and audio-language models
Requirements:
• Roughly 5+ years of hands-on industry experience in speech or audio ML
• A Bachelor's degree in computer science, electrical engineering, or a related technical or quantitative field is required
• Trained and evaluated speech or audio models yourself — ASR, TTS, speech-to-speech, speaker, or audio-language models — with strong PyTorch fundamentals
• Fluency in ESPnet, NeMo, SpeechBrain, Kaldi, HuggingFace, forced alignment, WER/CER, and related speech metrics
• Hands-on experience with multilingual, accented, dialectal, low-resource, or code-switched speech
• Experience with synthetic or augmented audio, including TTS pipelines, noise and room-response simulation
• Built evaluation sets and reasoned about dataset coverage across conditions
• First-author publications or strong open-source contributions at venues such as Interspeech, ICASSP, ASRU, SLT, or NeurIPS
• Ability to work directly with customer and frontier-lab research scientists
• Ability to explain data and modeling decisions clearly to expert and non-expert audiences
• Rigorous, reproducible approach to experiments and documentation
• Interest or hands-on experience in responsible-AI evaluation and red-teaming is a bonus
Benefits: