WellSaid Producer puts AI voiceover inside Claude and ChatGPT via MCP — native 48kHz audio
WellSaid launched WellSaid Producer 81 minutes ago, turning its professional voiceover studio into headless tooling that runs inside Claude, Copilot, ChatGPT or any agentic harness through a Model Context Protocol (MCP) server. The headline spec is audio generated natively at 48kHz — the broadcast standard — so finished voiceover drops straight into a video timeline with no upsampling.
What we checked: WellSaid’s announcement, published November 10, 2026 at 11:11 a.m. ET via PR Newswire, cross-checked against Unite.AI’s same-day coverage.
- What: WellSaid Producer — headless, composable voiceover production tooling that runs inside AI assistants through an MCP server, instead of only inside WellSaid Studio.
- When: announced November 10, 2026, at 11:11 a.m. ET; available now.
- Audio: native 48kHz generation — the standard for film, video and broadcast — with no upsampling needed. WellSaid says it is the first voiceover production platform to generate natively at 48kHz; 96kHz output for editing and mixing has been supported since 2025.
- Pronunciation: agentic style guides powered by a model trained on more than 500,000 Oxford Languages transcriptions.
- Access: works in Claude, Copilot, ChatGPT or any agentic harness via WellSaid’s MCP server; finished audio goes straight to a CMS, DAM or LMS. Demo via wellsaid.io/contact-sales; no pricing disclosed.
- Scale: WellSaid’s customers include nearly half of the Fortune 100, and the platform has produced more than 100 million minutes of professional voiceover since 2019.
Voiceover used to mean booking a studio. Then AI made the studio virtual. WellSaid’s new move is to take the studio out of its own website entirely and put it where the script is already being written — inside an AI assistant.
What Producer actually does#
Producer reimagines WellSaid Studio as headless, composable tooling. From Claude, Copilot, ChatGPT or a team’s own software, connected through WellSaid’s MCP server, a team can cast a licensed AI voice, direct the delivery, fix pronunciation, review takes and version the audio — then send the finished file straight to a content management system, digital asset manager or learning management system, without downloading and re-uploading anything.
Take the concrete case WellSaid is pitching: a course module. The script is drafted in Claude, the voice is cast and directed there, a product name that the model keeps mispronouncing gets fixed once in a style guide, and the finished narration lands in the LMS. No tab-switching to a voiceover app, no file juggling between tools.
“Voiceover was once recorded in a studio, and AI made the studio virtual. Now voiceover is part of every creative professional’s workflow,” said Benjamin Dorr, CEO of WellSaid. “Teams are remixing capabilities from every tool they use. Our job is to make WellSaid’s voices the professional production standard.”
The launch is also deliberately open. Teams can build WellSaid’s editing capabilities into their own tools, and — unusually for a voice vendor — use voice models from other providers alongside WellSaid’s in a single workflow. Customers can also build a library of custom digital voice replicas for their brand, with consent and licensing on record, the same way WellSaid’s standard voices are licensed.
The 48kHz number, in context#
The spec that matters most to video teams is the sample rate. Most AI voice tools generate at lower rates and upsample for delivery; WellSaid says Producer generates natively at 48kHz, the standard for film, video and broadcast, so audio drops into a timeline at full fidelity. The company claims to be the first voiceover production platform to do it. It has supported 96kHz output for commercial audio editing and mixing since 2025 — but that was output, not generation.
| Number | What it is |
|---|---|
| 48kHz | Native generation rate — broadcast standard, no upsampling |
| 96kHz | Output rate for editing and mixing, supported since 2025 |
| 500,000+ | Oxford Languages transcriptions behind the pronunciation model |
| 100M+ | Minutes of professional voiceover produced since 2019 |
| ~half | Share of the Fortune 100 among WellSaid’s customers |
The pronunciation model is the quieter differentiator. Style guides powered by a model trained on more than 500,000 Oxford Languages transcriptions are meant to keep brand names, product names, acronyms and technical terms pronounced correctly across every file — the kind of consistency failure that makes AI voiceover sound amateur at scale. For teams producing training or marketing audio in volume, a mispronounced product name in one video out of fifty is the failure mode; WellSaid is selling the fix as a feature, not a setting.
That licensing line is doing real work in the announcement. Every WellSaid voice comes from a professional actor who gave consent and is compensated when the voice is used — a contrast with voice-cloning tools that let anyone clone any voice. WellSaid’s voice-cloning rivals ship flexibility; WellSaid is shipping paperwork: only voices tied to specific individuals, the company argues, can clear rights of publicity, privacy and biometric identity. Voice actor Alicia Harris, one of the voices on the platform, said the new product “makes my voice even easier to use” while keeping actors “part of the process and compensated.”
The catch#
Three things temper the launch. First, there is no pricing anywhere in the announcement — Producer is “available now” behind a demo request, which makes it an enterprise conversation, not a self-serve signup. Second, the “first to generate natively at 48kHz” claim is WellSaid’s own; no independent benchmark was published with the launch. Third, the whole pitch rides on MCP: WellSaid calls MCP and agents “as disruptive to software as the Web once was,” but if teams do not actually direct voiceover from inside their assistants, Producer is a feature looking for a workflow. The voice-model race is also moving fast, and a headless studio still depends on the host assistant being good at the creative direction part.
None of that undercuts the core offer. For a team already producing video or training content at volume, casting and directing a licensed voice from inside Claude — and getting broadcast-rate audio back without a round trip through a separate app — removes a real step. The question is whether that step was the bottleneck, or whether the bottleneck is everything around it: approvals, brand review, and the fact that most teams still write scripts in a doc, not in a chatbot.