Skip to main content
TALENT247 // SPECIALIST AI EXECUTIVE SEARCH & TECHNICAL RECRUITMENT
RES · AI Research

Senior Audio ML Research Engineer, Speech-to-Speech Models

London, UK · Full-time · Hybrid

AI Specialism

AI Research

Workplace

Hybrid

Location

London, UK

Compensation

£125k – £175k

annual package

Role Dossier & Specification

REF: SENIOR-AUDIO-ML-RE

THE CLIENT & MANDATE

Our client is a flagship generative audio research lab and multi-billion-dollar unicorn with a major R&D center in central London, widely credited with defining the state of the art in expressive human-level speech synthesis, voice cloning, and real-time voice interaction. Their next frontier moves beyond cascaded ASR-to-LLM-to-TTS pipelines toward unified, native speech-to-speech foundation models that reason directly over interleaved audio and text tokens—capturing laughter, hesitation, emphasis, back-channeling, and fluid duplex overlap without rigid turn-taking.

They have retained Talent247 to appoint a Senior Audio ML Research Engineer, Speech-to-Speech Models in London. Working inside a small, high-autonomy R&D team alongside world-class speech and multimodal researchers, you will design, pre-train, and post-train end-to-end conversational audio foundation models shipped to millions of users and developers globally.

KEY RESPONSIBILITIES

  • Propose, implement, and scale novel interleaved audio-text autoregressive, flow-matching, and discrete/continuous neural audio codec architectures for real-time full-duplex speech-to-speech generation.
  • Train ultra-low-frame-rate, semantic-rich neural audio tokenizers balancing acoustic fidelity, speaker identity preservation, paralinguistic expressiveness, and causal streaming latency (<80ms lookahead).
  • Lead post-training alignment (DPO, RLHF, distillation) for spoken conversational models—optimizing instruction following, emotional appropriateness, natural interruption handling, and voice consistency across 30+ minute dialogues.
  • Define new latency-aware and interaction-aware evaluation frameworks measuring duplex naturalness, paralinguistic accuracy, and spoken reasoning retention.
  • Curate and filter hundred-thousand-hour multi-channel conversational audio corpora with rich paralinguistic and speaker-turn annotations.
  • Partner closely with CUDA/inference engineers to ship quantized streaming models onto production GPU fleets.

TECHNICAL ENVIRONMENT & OFFERING

Researchers work in Python and PyTorch across dedicated multi-node H100/B200 clusters with custom distributed audio data loaders and fast production deployment paths. Based hybrid in London, our client offers £125,000–£175,000 GBP base salary plus high-upside unicorn equity, comprehensive UK benefits, and full Global Talent / Skilled Worker visa sponsorship.

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 5+ years of ML research and engineering experience in generative speech synthesis (TTS), neural audio codecs, speech language models, or multimodal audio-text modeling.
  • Expert proficiency in PyTorch, distributed multi-node training (FSDP/Megatron), causal streaming transformers, flow matching/diffusion, or RVQ/FSQ discrete tokenizers.
  • Proven track record taking novel generative audio or sequence architectures from initial hypothesis and dataset assembly all the way to shipping-quality checkpoints.
  • Deep intuition for time-series/acoustic modeling, receptive fields, streaming lookahead constraints, and audio perceptual quality tradeoffs.
  • Strong general software engineering discipline within a shared, fast-moving research codebase.

PREFERRED / BONUS EXPERIENCE

  • PhD in Machine Learning, Speech/Audio Processing, or Computer Science with publications at Interspeech, ICASSP, NeurIPS, ICML, or ICLR.
  • Direct hands-on experience training full-duplex speech-to-speech models or distilling autoregressive/flow audio models for sub-100ms streaming inference.

Position Telemetry

Location London, UK (Hybrid)
Department AI Research
Employment type Full-time
Workplace type Hybrid
Salary range
£125k – £175k annual package