OpenRouter
for Speech.

Discover the models that listen, speak, and translate.

Explore speech tools Try the playground ↓

Startup programs & cloud support

Speech to text

Explore tools

Provider to explorePrepared sample

Audio inputMP3

Team catch-upHindi + English

0:00 / 0:03

Read source transcript

Meeting kal 3 baje hai. Please send the notes.

Sample output

Original languages preserved

Meeting kal 3 baje hai. Please send the notes.

Illustrative output

Samples are illustrative, independent of provider.Deepgram docs

Sample modeNo live API requests · no keys required

Your speech stack, with room to grow.

Explore all 12 tools

A whole world of speech.
A place to find your fit.

DeepgramElevenLabsCartesiaValsea+ more

Explore providers in the directory. Listings are not active Humlet integrations.

What should your
agent do next?

Start with a capability. Find the tools behind it.

Browse all 12 tools

Listen closely.

Speech to text for live conversations and recordings.

Transcripts · captions · speaker labels

Say it naturally.

Text to speech for an agent with something to say.

Voice agents · narration · audio

Find common ground.

Speech translation that brings languages together.

Meetings · conversations · accessibility

The idea behind Humlet

More conversation.
Less connecting the dots.

Speech is a stack of decisions. We’re building one place to connect them—starting with discovery, then a shared integration layer.

Your AI agentYour application

humletOne speech layer

Speech to textText to speechTranslation

Explore now

Find your speech stack.

Compare tasks and audio modes across 8 providers, with links to their documentation.

In development

Connect through Humlet.

Our direction: a consistent speech interface, starting with transcription and translation.

The longer view

Keep your options open.

Provider choice and routing without rebuilding your application around every new model.

For the people building it

Start with the
conversation.
Then choose the tools.

A voice agent and a meeting companion don’t need the same speech stack. Map your needs before choosing a model.

Pick the speech tasks your product needs.
Evaluate languages, latency, and output quality.
Bring a concrete brief to an integration conversation.

I’m building

{
  "project": "A voice agent",
  "capabilities": [
    "speech-to-text",
    "text-to-speech"
  ],
  "audio": "streaming",
  "evaluate": [
    "turn-taking",
    "first-audio latency",
    "voice quality"
  ],
  "provider_preference": "To be decided",
  "access": "Request Humlet early access"
}

A planning brief. No SDK or credentials required.

ChirpberryApp preview

Spanish

Las buenas ideas
no tienen fronteras.

English

Good ideas
have no borders.

A conversation, with room for everyone.

Meet the companion app

A speech layer.
A human side.

Chirpberry brings the story into an application: a companion for conversations that move between languages.

Chirpberry is the app, powered by Hushfig. Humlet is our separate direction for discovering and connecting speech tools.

Meet Chirpberry

A few things
worth asking.

Is this OpenRouter for speech?

Humlet is an independent project inspired by that idea: a simpler way to discover and eventually access speech capabilities through one layer. Today, this site offers a curated directory. Multi-provider routing and unified billing are not yet available through Humlet.

Can I use every tool through Humlet today?

Not yet. The directory describes services available from their respective providers, with links to official documentation. A listing is not an active integration or a partnership. Humlet developer access is being prepared.

Does it work with every language and accent?

Coverage depends on the provider, model, and task. Evaluate your actual languages, accents, noise conditions, and terminology. The directory helps you find candidates; it does not claim universal recognition or translation quality.

Can I get an API key or try live audio?

Self-service keys and live audio processing are not available on this website. The playground uses prepared audio and illustrative outputs, independently of the selected provider. You can explore the directory, prepare an integration brief, and visit the Chirpberry application website.