Why I built Murmur: a $49 on-device alternative to ElevenLabs

Every text-to-speech tool I tried for my own projects had the same three problems: pay-per-character pricing that turns long-form work into a tax, mandatory cloud upload of whatever I was narrating, and a subscription that kept charging even when I went a month without using it.

So I built Murmur (https://murmurtts.com/) for myself. Then I made it into a product.

WHAT MURMUR DOES

Murmur is a native macOS text-to-speech studio. You paste text, pick a voice, hit Generate, and it outputs a WAV or M4A file. Everything happens on your Mac using Apple's MLX framework. No cloud calls, no telemetry, no per-character fees.

- 860+ voices across 25+ languages

- Voice cloning from a 10-second audio sample

- Four AI models in one app: Kokoro, Qwen3-TTS, Fish S2 Pro, and Chatterbox

- Batch processing for long-form work like audiobooks

- 3x real-time generation on Apple Silicon

- $49 one-time, no subscription, no usage caps

See live samples: https://murmurtts.com/samples

WHO IT'S FOR

Murmur is built for creators who actually ship audio: podcasters writing intros and ads, audiobook authors narrating manuscripts, YouTubers producing voiceovers, newsletter writers turning posts into audio editions. Basically anyone whose cloud TTS bill scales with their output.

The on-device part matters more than I expected. Lots of buyers tell me they were nervous about uploading client manuscripts, internal training scripts, or unpublished work to a third party. Local generation removes that friction entirely.

WHY ON-DEVICE, WHY NOW

Apple Silicon plus MLX changed what's possible on a Mac. Five years ago you couldn't run a serious TTS model locally at usable speed. Today an M-series Mac runs four production-grade models at 3x real-time without breaking a sweat.

That capability didn't really exist when ElevenLabs and Murf were built. Their pricing was the only model that worked when inference had to happen in someone else's data center. Now there's a real alternative, and on-device economics let me charge once instead of metering every paragraph.

WHAT'S NEXT

Active areas I'm working on for Murmur:

- More voices in underrepresented languages

- Faster cold start

- Tighter export formats for podcast and audiobook workflows

- A proper voice library UI for managing cloned voices

If you're shipping audio content on a Mac and want to compare notes, grab Murmur at https://murmurtts.com/ and let me know what's missing for your workflow. I read every email.

Try it: https://murmurtts.com/

Sample voices: https://murmurtts.com/samples

Use cases: https://murmurtts.com/use-cases