About Hafsati
Je suis docteur en traitement du signal audio & IA, et j’interviens comme Lead Audio sur des chaînes audio complètes (du capteur jusqu’au cloud) : robustesse bruit/accents, latence, coûts, privacy, monitoring qualité.
J’aide les équipes Produit/Data/Tech à décider vite et bien sur des sujets voix (ASR, diarisation, audio analytics, quality monitoring) et à passer de POC à production avec une approche mesurable (métriques, protocole d’évaluation, analyse d’erreurs, roadmap).
Ce que vous gagnez : clarté + réduction du risque + plan d’exécution concret (edge/cloud, on-prem ou SaaS).
Expertise : audio ML / speech, évaluation & benchmarking, data readiness/annotation, architecture edge+cloud, MLOps/monitoring, optimisation qualité/coûts/latence.
Offres packagées:
- Audit IA Voix + Roadmap (10 jours ouvrés)
- Benchmark & Vendor Due Diligence (7 jours ouvrés)
French
Native or bilingual
Arabic
Native or bilingual
English
Fluent
Can work on-site
Paris (up to 50km)
Experience
- Enchanted ToolsAudio Technical Specialist and AI expertDecember 2023 - Today (2 years and 6 months)Paris, FranceDevelop and deploy advanced audio signal processing pipelines to enhance the robot’s auditory perception (robust front-end processing, noise robustness, and real-time constraints).Build AI-driven audio capabilities including speech recognition (ASR), sound event understanding, and sound localization to support natural human–robot interaction.Implement and optimize on-device ASR using Whisper and Wav2Vec 2.0, targeting low-latency inference on edge hardware.Design and deploy Text-to-Speech (TTS) and speech transcription systems, including flow-matching-based TTS optimized for Jetson Orin / Xavier.Integrate audio services across the robot stack using gRPC-based communication, enabling reliable streaming, modularity, and scalable deployment.Optimize end-to-end performance on embedded platforms (profiling, quantization/acceleration where applicable, memory/latency trade-offs) to meet real-time requirements.Deliver end-to-end “live” conversational pipelines by integrating streaming ASR + TTS with Gemini Live, enabling low-latency interactive voice experiences.Collaborate with cross-functional robotics teams (perception, embedded, product) to ensure seamless integration, robust testing, and production readiness.Stay current with state-of-the-art speech and audio research, rapidly prototyping and transferring new methods into production.
- TUITOTechnical Referee Audio Signal Processing and AIMarch 2021 - November 2023 (2 years and 8 months)13600 La Ciotat, FranceBuilt real-time Voice Activity Detection (VAD) pipelines to reliably segment speech from noise, improving responsiveness and overall UX in production audio streams.Designed and integrated wake-word detection models optimized for high precision and low false-activation, ensuring robust always-on triggering across devices.Developed speech enhancement solutions using advanced signal processing and learning-based methods to reduce noise/reverberation and improve intelligibility in challenging acoustic conditions.Implemented speech separation and sound source separation techniques to isolate overlapping speakers and sources in multi-speaker, noisy environments.Developed and fine-tuned multilingual Automatic Speech Recognition (ASR) systems, leveraging deep learning architectures and large-scale dataset adaptation to boost accuracy.Built Natural Language Understanding (NLU) components for conversational systems, improving intent classification and entity extraction for reliable query interpretation.Delivered Speech-to-SQL capability: mapped spoken queries into structured SQL for voice-driven database retrieval and manipulation.Designed and deployed conversational AI agents with contextual understanding and multi-turn dialogue handling, improving naturalness and task completion.
- Sorbonne Paris North University (Université Sorbonne Paris Nord) & Université de ToulonTeaching Assistant (Signal Processing & Machine Learning / Robotics)TECHSeptember 2021 - August 2025 (3 years and 11 months)Paris, FranceDeliver tutorials (TD) and practical labs (TP) for Bachelor’s (L3SPI) and Master’s programs (M1 Image & Networks, Master ROC – Robotics).Teach Signal Theory and Digital Signal Processing, helping students connect mathematical foundations to real-world engineering applications.Design and deliver courses on Python fundamentals for robotics and data-driven engineering workflows.Explain and supervise supervised learning concepts: perceptrons, MLP, forward pass/loss computation, backpropagation, and optimization via gradient descent.Teach deep learning models and modern architectures: CNNs, RNNs, autoencoders, attention mechanisms, and transformers.Lead sessions on multimodal perception, combining image/audio/text data for robust perception and analysis.Adapt pedagogy to student level and learning needs, emphasizing critical thinking, structured problem-solving, and hands-on implementation.Collaborate with academic teams to develop and improve course materials and ensure learning-outcome quality.
Recommendations
Be the first to recommend Hafsati
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Doctor of Philosophy - PhDUniversité de Rennes 12020Doctor of Philosophy - PhD
- Master's Research Signal Image System EmbArqué (SISEA)Supélec Rennes, Université de Rennes 12016Master's Research SISEA