Applied Scientist Intern (Audio Language Modeling)
Reality DefenderSalary not specified
- Full Time
- Internship
$150,000 – $220,000Compensation
Mid-level (3 to 4 years), Senior (5 to 8 years)Experience Level
Full TimeJob Type
UnknownVisa
AI & Machine Learning, Data & AnalyticsIndustries
Requirements
Candidates must possess a strong mathematical foundation in statistical learning theory, particularly in self-supervised and multimodal learning. They should have deep expertise in foundation model architectures and the ability to scale training across multiple modalities. Proven ability to bridge theory and practice is essential, along with experience in building data pipelines for processing and curating massive datasets. A track record of designing controlled experiments and optimizing models for real-world deployment is required, as well as a history of open-source contributions or research publications in speech and language AI.
Responsibilities
The Research Scientist will pioneer the development of Latent Space Models aimed at addressing data, scale, and cost challenges in building robust voice AI. They will build next-generation neural audio codecs, develop steerable generative models for human speech synthesis, and create embedding systems for codec latent space interpretation. Additionally, they will leverage latent recombination to generate synthetic audio data, train multimodal speech-to-speech systems, and design model architectures, training schemes, and inference algorithms for efficient training and real-time inference.
Skills
Latent Space Models
Neural Audio Codecs
Speech-to-Text
Text-to-Speech
Speech-to-Speech
Audio Processing
Machine Learning
Deep Learning
Python
PyTorch
TensorFlow
Deepgram
Speech recognition APIs for audio transcription
About Deepgram
Deepgram specializes in artificial intelligence for speech recognition, offering a set of APIs that developers can use to transcribe and understand audio content. Their technology allows clients, ranging from startups to large organizations like NASA, to process millions of audio minutes daily. Deepgram's APIs are designed to be fast, accurate, scalable, and cost-effective, making them suitable for businesses needing to handle large volumes of audio data. The company operates on a pay-per-use model, where clients are charged based on the amount of audio they transcribe, allowing Deepgram to grow its revenue alongside client usage. With a focus on the high-growth market of speech recognition, Deepgram is positioned for future success.
Key Metrics
San Francisco, CaliforniaHeadquarters
2015Year Founded
$100.5MTotal Funding
SERIES_BCompany Stage
Data & Analytics, AI & Machine LearningIndustries
51-200Employees