Eleven roles this week, and most of them come from big tech. Before anything else, one deadline: Zoom's audio deepfake detection role closes October 28, the day after this goes out. Google, Amazon, Microsoft and Salesforce each posted a senior speech-model role, and Salesforce pays the most: a director who will run the Agentforce Voice Models team, with a base of up to $344.7K. If you don't have a PhD, there are still good options. Cresta and Deepgram are both hiring remotely in the US, and Hippocratic AI wants a linguist to own the quality of the transcription data behind its healthcare voice agents.
Speech & audio research
- Audio AI Engineer — Zoom
· San Jose, CA (onsite 5 days) · $137.7K–$275.4K + bonus + equity · applications close Oct 28
Builds models that detect audio deepfakes and synthetic speech on Zoom's applied research team. You'd design the training datasets and evaluations, make detection robust across speakers, languages, devices and codecs, and take prototypes to production. Asks for a PhD or equivalent experience, plus research or industry work in deepfake detection, anti-spoofing or speaker recognition. Voice cloning and adversarial ML are pluses. Apply now if you're interested: the posting comes down tomorrow. View posting - Staff Research Scientist, Foundational Audio — Google
· Mountain View, CA · $174K–$252K + 15% bonus target + equity
Sets the research agenda for sequence, speech and audio modeling at Google Research, and co-designs models for scaling, memory and cache efficiency, distillation and low-latency streaming inference. Requires a PhD or equivalent, 2 years leading a research agenda in this area, and published papers. You also need hands-on work with at least one of state-space or linear attention, flow matching, speculative decoding or streaming alignment. Contributions to Kaldi, K2, ESPnet or Hugging Face are listed as pluses. View posting - Senior Applied Scientist, Real-Time Conversational AI — Amazon (AGI)
· Sunnyvale, CA ($192.2K–$260K) or Boston, Bellevue or Seattle ($167.1K–$226.1K) · + sign-on + RSUs
Trains large multimodal foundation models for real-time speech and audio generation. You'd also build the reward models and RL that teach them natural timing and conversational behavior. Requires a PhD, or a master's plus 6+ years, along with hands-on foundation-model training and papers at venues like Interspeech, ICASSP or NeurIPS. We listed this title on September 29. That posting has since come down, and this one adds Boston, Bellevue and Seattle to Sunnyvale. View posting - Senior Applied Scientist, CoreAI Voice Agent — Microsoft
· Redmond, WA (3 days/week in office) · $119.8K–$234.7K ($160.2K–$261K in the Bay Area and NYC)
Develops natural, expressive, multilingual speech generation for Microsoft's voice agent team. The job spans new algorithms, deployment to production, live comparisons between models, and monitoring of what's shipped. You need a bachelor's plus 4+ years, a master's plus 3+, or a PhD plus 1+. You must also pass the Microsoft Cloud background check. Posted October 21. View posting
Speech infrastructure & model engineering
- Senior Inference Engineer, AGI — Amazon
· Sunnyvale, CA ($193.3K–$261.5K) or Boston or Seattle ($168.1K–$227.4K) · + sign-on + RSUs
Serves frontier-scale speech and audio models within real-time latency budgets. That covers co-designing architectures with scientists, quantization, speculative decoding and custom kernels, continuous batching, and the rollout and reward serving behind RL. Requires 5+ years of software development, experience as a tech lead, and 2+ years profiling and improving inference frameworks, not just using them. vLLM or TensorRT-LLM internals and CUDA or Triton kernels are pluses. This is a second opening for the role. The req we listed on September 8 is still live. View posting - Member of Technical Staff, Multi-Modal Audio — Liquid AI
· Boston (hybrid) · compensation not listed (base + equity)
Works directly with the technical lead of Liquid's audio team, which builds a single on-device model that handles speech-to-text, text-to-speech and speech-to-speech. You'd own audio data pipelines (preprocessing, augmentation, quality filtering), evaluation systems and customer fine-tuning work. Requires production-grade code, shipped ML systems beyond training, and PyTorch with DeepSpeed or FSDP. Audio experience is a plus, not a requirement. This is a different opening from the post-training role we listed on October 13. View posting - Sr. Backend Engineer, Engine Team: API — Deepgram
· Remote (US) · $150K–$220K + 10% bonus + equity
Works on Deepgram's core inference services: networking, speech processing, audio transcoding, and latency and memory optimization. You'd debug scheduling and HPC problems and customize backend services for customers. Requires 3+ years in industry, Rust (or C/C++) with competent Python, and deep experience with UNIX-style systems. ML and audio processing experience are pluses. View posting
Speech data quality
- Audio Transcription Specialist — Hippocratic AI
· Menlo Park, CA (onsite 5 days) · compensation not listed
Owns the quality of the transcription data behind the healthcare voice agents of a company valued at $3.5B. You'd review output from outside transcription vendors and give them feedback, set up QA workflows, validate audio pipelines, and keep up an error taxonomy that researchers use to decide which ASR problems to fix first. Requires a bachelor's in linguistics, speech and language sciences or a related field, plus 2+ years of transcription, annotation or audio QA. Knowing WER and having evaluated accented or noisy speech are pluses. Unusually for this list, you don't need a PhD or an engineering background. View posting
Voice agents & voice platforms
- Director of Product Management, Agentforce Voice Models — Salesforce
· San Francisco or Bellevue, WA (flexible) · $197.3K–$313.7K ($237.7K–$344.7K in select SF and NYC areas)
The title says product management, but this person owns the entire voice model stack: ASR, TTS, speech-to-speech, diarization, voice biometrics and language coverage. They also set the WER, MOS and latency targets, make the build-or-buy calls, and hire a team of 20 to 30 research engineers and scientists. Requires 10+ years in speech or audio ML, including 4+ years managing teams of 10 or more. You also need deep hands-on work in at least two of ASR, neural TTS or real-time speech pipelines, and production models running at millions of minutes a day. View posting - Lead Engineer, AI Agent Voice Experience — Cresta
· Remote (US) · $205K–$270K + equity
Leads work on how Cresta's contact-center AI agents listen and talk, at a company whose platform also covers conversation intelligence. That covers ASR, turn detection, TTS and speech-to-speech, plus the evaluation frameworks and failure analysis behind them. It also covers caller-facing details like backchanneling, focusing on the main speaker, and responding with empathy. Requires 5+ years deploying ML in production, experience leading technical teams, and a background in speech, NLP or voice AI. WER and contact-center experience are pluses. View posting - Agentic AI Developer — Technatomy
· Listed as remote, but see note · compensation not listed · U.S. citizen or green card holder; Public Trust
A federal contractor modernizing IVR for an enterprise health customer with LLM voice agents. The work covers STT and TTS, RAG, tool calling and human escalation on Amazon Connect, Lex or Genesys Cloud. Requires 12+ years in software or enterprise integration. The header says remote, but the body names Baltimore, MD, and prefers candidates who can be onsite about 75% of the time, so confirm with the recruiter. View posting
Hiring for a speech-tech role? Tell us about it and we'll include it here and in the digest. Browse everything by specialty on the homepage.