Director of Product Management (Agentforce Voice Models)
Posted Updated 6
views
Who recruits here Pro
Up to three recruiters at Salesforce with their work email, so you can write to a person.
Part of Pro. How it works
Description
- 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
Benefits
- Medical Care
- Life Insurance
- Retirement Savings
- Employee Assistance Programs
- With 9 standard holidays and four floating holidays, you get a total 13 paid days off each year
- 10+ years in speech/audio machine learning, with 4+ years in a senior leadership role managing teams of 10 or more engineers or scientists
- 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)
- Excellent written and verbal communication—able to translate complex model behavior into product and business narratives for executives
- 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)
- 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
- PhD in Computer Science, Electrical Engineering, Linguistics, or related field with a speech/audio focus, or equivalent industry experience
- Experience with voice safety, watermarking, and deepfake/spoofing detection
- Familiarity with streaming inference, model quantization (INT8/INT4), and on-device deployment for low-latency voice
- Background in telephony protocols (SIP, RTP, WebRTC) and enterprise contact-center platforms (Genesys, NICE, Avaya, Amazon Connect)
- Published research or patents in speech processing, audio ML, or a closely related field
- Experience operating within regulated industries (financial services, healthcare) with voice data compliance requirements
Skills
Sign in to read and join the discussion.