Senior Speech & NLP ML Engineer

Svitla Systems Inc. is looking for a Senior Speech, Audio & NLP ML Engineer or a full-time position (40 hours per week) in Ukraine, Turkey. Our client is an American mobile computing company.

The ideal candidate combines strong software engineering expertise with hands-on experience in speech recognition, audio processing, and NLP. We’re looking for someone who enjoys solving complex technical challenges, takes full ownership of their work, and continuously looks for ways to improve model performance and overall system efficiency.

You will work on developing, optimizing, and improving end-to-end speech processing pipelines, from raw audio input to accurate, structured text output.

Requirements

• Strong Python programming skills.
• Strong problem-solving and analytical abilities.
• Solid knowledge of machine learning and artificial intelligence.
• Good understanding of speech processing, audio processing, and NLP concepts.
• Hands-on experience with automatic speech recognition or speech-to-text systems.
• Experience building or improving audio preprocessing and transcription pipelines.
• Experience with audio data handling, including formats, sampling rates, noise, segmentation, and audio quality issues.
• Experience preparing and managing audio/text datasets for training, evaluation, and analysis.
• Experience fine-tuning LLM and/or ASR models for task-specific or domain-specific use cases.
• Experience evaluating ASR systems using metrics such as WER, CER, and task-specific quality metrics.
• Experience performing ASR error analysis and identifying ways to improve transcription quality.
• Experience with transcript post-processing tasks such as text normalisation, punctuation restoration, casing, or domain-specific vocabulary correction.
• Familiarity with deep learning frameworks such as PyTorch or similar.
• Familiarity with speech/audio and ML tooling such as Hugging Face Transformers/Datasets, torchaudio, librosa, pydub, soundfile, Pandas, NumPy, or similar.
• Experience with development and experiment tools such as Git/GitHub, Hydra, and Weights & Biases, or similar alternatives.
• Experience working with shared codebases, remote teams, and remote servers.
• Ability to independently own tasks, analyse performance, and proactively propose improvements.

Will be a plus

• Experience with Whisper, wav2vec 2.0, NeMo, Kaldi, SpeechBrain, or similar speech frameworks/models.
• Experience fine-tuning ASR models for specific domains, languages, accents, or noisy environments.
• Experience with multilingual or low-resource ASR.
• Experience with audio denoising or speech enhancement.
• Experience with voice activity detection and advanced audio segmentation.
• Experience with speaker diarisation, speaker verification, speaker identification, or speaker embedding models.
• Experience with streaming or real-time ASR systems.
• Experience with text-to-speech or voice synthesis is a plus, but not required.

Responsibilities

• Develop, train, evaluate, and deploy machine learning models for speech and audio-related tasks.
• Build and optimise end-to-end speech-to-text pipelines, including audio preprocessing, ASR, and transcript post-processing.
• Prepare, clean, manage, and manipulate audio and text datasets for model development and evaluation.
• Perform audio preprocessing tasks such as resampling, normalisation, denoising, segmentation, and format conversion.
• Analyse ASR outputs and identify ways to improve transcription quality.
• Improve ASR performance through data improvements, model tuning, pipeline optimisation, and post-processing.
• Apply NLP techniques for transcript cleanup, text normalisation, punctuation restoration, casing, and domain-specific correction.
• Evaluate speech and NLP systems using metrics such as WER, CER, precision, recall, and task-specific quality metrics.
• Run and track experiments, document findings, and communicate model or pipeline improvements clearly.
• Collaborate with team members on shared codebases, remote projects, and remote servers.