Director of Product Management, Salesforce

• 10+ years in speech/audio machine learning, with 4+ years in a senior leadership role managing teams of 10 or more engineers or scientists
• Deep hands-on expertise in at least two of: ASR (end-to-end or hybrid CTC/attention architectures), neural TTS (VITS, Voicebox, Matcha-TTS, or equivalent), or real-time speech processing pipelines
• Proven track record of shipping production voice models at scale (millions of minutes per day) with measurable accuracy and latency improvements
• Fluency in prosody modeling: pitch contour modeling, duration prediction, and expressiveness fine-tuning
• Experience with speaker diarization systems (clustering-based, end-to-end, or hybrid)
• Strong understanding of multilingual and multi-dialect NLP/ASR challenges; experience with at least 5 languages in production
• Proficiency in Python; experience with PyTorch or JAX model training at scale; familiarity with cloud-based training infrastructure (AWS, GCP, or Azure)
• Excellent written and verbal communication—able to translate complex model behavior into product and business narratives for executives

Desirable

• Agentforce is Salesforce's next-generation AI platform, delivering autonomous agents that reason, take action, and communicate naturally across every customer touchpoint
• Voice is the fastest-growing interaction surface—from contact-center automation and field-service assistants to real-time sales coaching and multilingual global deployments
• As Director of Agentforce Voice Models, you will own the full voice intelligence stack: Automatic Speech Recognition (ASR / STT), Text-to-Speech (TTS), Speech-to-Speech (S2S) end-to-end pipelines, speaker diarization, prosody modeling, and the language-coverage roadmap that makes Agentforce sound natural in every market we serve
• You will partner with product, infrastructure, and go-to-market teams to set the bar for transcription accuracy, latency, and voice expressiveness at enterprise scale
• Voice Model Strategy & Roadmap:
• Define and own the multi-year roadmap for Agentforce voice capabilities, spanning ASR/STT, TTS, S2S, and real-time voice agents
• Set accuracy, latency, and quality benchmarks (WER, MOS, RTF, DMOS) and drive the organization to meet them
• Evaluate build vs. buy vs. partner decisions for new voice model capabilities and maintain relationships with key academic and industry partners
• ASR / STT (Automatic Speech Recognition):
• Lead the development of production-grade ASR systems optimized for telephony, WebRTC, and device-side deployment.Word Error Rate (WER) improvement across noise conditions, accents, and domain-specific vocabularyStreaming and batch recognition pipelines with sub-200 ms first-token latency targetsCustom vocabulary and language model adaptation (hot-word boosting, domain LM interpolation)Punctuation restoration and inverse text normalization (ITN) for downstream NLU
• Drive multilingual and code-switching ASR coverage across priority languages; govern the language onboarding process including data acquisition, model training, and acceptance testing
• TTS (Text-to-Speech):
• Own the neural TTS pipeline—voice cloning, persona design, SSML compliance, and real-time synthesis—for Agentforce agent personas
• Lead prosody research: intonation, rhythm, stress, and pause modeling that produces natural-sounding enterprise voices across conversational contexts
• Manage voice talent agreements, ethical AI review, and consent frameworks for synthetic voice creation
• Drive naturalness, expressiveness, and brand-consistency quality bars using subjective (MOS, CMOS) and objective (mel-cepstral distortion) evaluation frameworks
• Speech-to-Speech (S2S) & Real-Time Voice Agents:
• Architect low-latency S2S pipelines that enable full-duplex conversational AI without the ASR→NLU→TTS handoff penalty
• Partner with the Agentforce Reasoning team to integrate voice understanding with agent action loops (tool calls, CRM lookups, escalation routing)
• Establish interruption, barge-in, and turn-taking models appropriate for enterprise voice agents
• Speaker Diarization & Voice Analytics:
• Deliver production speaker diarization ("who spoke when") for multi-party calls, enabling accurate per-speaker transcripts used in call coaching, compliance, and analytics
• Develop speaker verification and voice-biometric capabilities for secure agent authentication use cases
• Partner with the Einstein Analytics team to surface voice-derived signals (sentiment, engagement, talk-time ratios) in Salesforce dashboards
• Language Coverage & Localization:
• Own the global language support roadmap; prioritize languages by customer demand, addressable market, and data availability
• Establish data governance, annotation, and quality-control pipelines for low-resource languages
• Work with regional Salesforce teams to validate dialect, accent, and cultural appropriateness of voice personas
• Engineering Leadership & Team Building:
• Recruit, develop, and retain a world-class team of research engineers, applied scientists, and ML engineers (target team size: 20–30)
• Set technical direction, drive architectural decisions, and maintain engineering excellence through code review culture, rigorous evaluation frameworks, and production incident reviews
• Build a culture of experimentation: rapid A/B testing, red-teaming for voice safety, and continuous model refreshes
• Cross-Functional Partnership:
• Partner with Product Management to translate customer and regulatory requirements into model specifications and acceptance criteria
• Collaborate with Legal, Privacy, and Trust & Safety on voice AI ethics, speaker consent, deepfake-detection, and compliance (HIPAA, PCI, GDPR)
• Represent Agentforce Voice at external conferences, standards bodies (W3C, IETF), and customer briefings