описание
We are looking for an experienced AI/ML Engineer specializing in Speech AI to develop, fine-tune, and optimize Speech-to-Text (STT) and Text-to-Speech (TTS) models. The primary focus of this role is improving speech recognition accuracy and speech synthesis quality, with a particular emphasis on supporting the Georgian language. You will work with open-source and pre-trained speech models, prepare and process audio datasets, conduct fine-tuning experiments, and optimize models for reliable production deployment.
задачи
- Fine-tune and optimize Speech-to-Text (STT) models.
- Develop and improve Text-to-Speech (TTS) models and natural voice quality.
- Prepare, clean, annotate, and validate audio datasets.
- Work with Georgian-language speech datasets and multilingual audio.
- Experiment with different speech model architectures and training techniques.
- Improve speech recognition accuracy, pronunciation, and language coverage.
- Evaluate models using metrics such as WER, CER, and MOS.
- Optimize models for inference speed, accuracy, and computational efficiency.
- Work with GPU-based training and inference environments.
- Research new speech AI architectures, techniques, and open-source technologies.
- Deploy, monitor, and maintain speech models in production environments.
требования
- Strong Python programming skills.
- Solid understanding of Machine Learning and Deep Learning.
- Practical experience with PyTorch or TensorFlow.
- Experience with Hugging Face and speech model training workflows.
- Understanding of STT and TTS architectures.
- Experience with audio processing, preprocessing, and dataset preparation.
- Knowledge of speech recognition and synthesis evaluation methods.
- Familiarity with GPU computing and CUDA.
- Experience with model fine-tuning, optimization, and inference.
- Understanding of audio signal processing fundamentals.
- Будет плюсом: Experience with NVIDIA NeMo; Experience fine-tuning models such as Whisper, FastConformer, or other speech recognition architectures; Experience with TTS models such as VITS, XTTS, or other neural speech synthesis architectures; Experience with voice cloning and speaker adaptation; Knowledge of phonetics, linguistics, and pronunciation modeling; Experience working with Georgian or other low-resource languages; Experience with distributed training and multi-GPU environments; Familiarity with Docker, FastAPI, and model serving
условия
- Opportunity to work on real-world Speech AI development.
- Opportunity to build and improve proprietary STT and TTS models.
- Work with modern open-source AI technologies and GPU infrastructure.
- Flexible working environment.
- Opportunity to influence the technical direction of our Speech AI products.