Ted Hisokawa
Mar 06, 2026 12:43
ElevenLabs deploys new generative mannequin letting customers design totally new artificial voices from scratch, focusing on audiobooks, video games, and content material creators.
ElevenLabs has deployed a generative AI mannequin that creates totally new artificial voices from scratch, addressing what the corporate calls a “severely underhyped” section of the AI market. The Voice Generator software lets customers design customized voices by setting parameters together with gender, age, accent, pitch, and talking model.
The function, rolling out by way of the corporate’s Voice Lab, generates distinctive voices with every use—even when an identical base parameters are chosen. This solves a sensible drawback: ElevenLabs discovered its present speaker financial institution too restricted for customers who wanted unique voices for his or her tasks.
How It Works
The technical strategy emerged from ElevenLabs’ present speech synthesis and voice cloning infrastructure. Each processes depend on speaker embeddings—vector representations that encode a voice’s traits. By coaching a devoted mannequin to pattern from the distribution of those embeddings, the corporate can now generate infinite variations.
The conditioning layer provides management. Customers aren’t simply rolling cube on random outputs; they’re specifying core identification markers that form the generated voice.
Goal Functions
The corporate is positioning the software throughout a number of verticals:
Publishing: Ebook authors can convert textual content to audio whereas sustaining inventive management over narration design—doubtlessly increasing the audiobook market to titles that could not justify conventional recording prices.
Information Media: Publishers experimenting with audio content material can create distinctive, unique voices for his or her manufacturers. The exclusivity angle issues right here—a voice representing one outlet will not present up elsewhere.
Sport Growth: Studios can voice NPCs that will in any other case stay silent, with voices distinctive to their digital worlds. The price-efficiency argument is easy: extra voiced content material with out proportional finances will increase.
Promoting: Creatives can prototype a number of voice types immediately throughout early marketing campaign improvement, earlier than committing sources.
Business Context
The launch arrives as voice AI advances quickly throughout the sector. Late 2024 noticed Azure launch its gpt-4o-mini-tts mannequin, whereas early 2026 introduced the open-sourced Qwen3-TTS household emphasizing voice design and multilingual streaming. The broader pattern factors towards orchestrated speech methods combining speech-to-text, massive language fashions, and text-to-speech—plus rising speech-to-speech fashions that bypass textual content conversion totally.
ElevenLabs can be telegraphing its subsequent transfer: combining voice era with voice cloning to let customers improve their very own voices. The pitch entails manipulating cloned voices to sound extra pure or different—focusing on anybody who data displays or audio messages however dislikes how they sound.
Security Measures
The corporate outlined a number of safeguards in opposition to misuse: phrases prohibiting unlawful or dangerous purposes, watermarking to hint generated audio again to the platform, and assessment processes for reported infringements. On the financial displacement concern, ElevenLabs argues voice actors may license their voices for AI coaching whereas collaborating in additional tasks concurrently.
Whether or not that framing satisfies working voice actors stays an open query as artificial voice high quality continues approaching human parity.
Picture supply: Shutterstock
