Decibel Labs / Model

Reso 1v1.0

Our first speech model. A streaming text-to-speech system that sounds human in English and Hindi, code-switches mid-sentence, and starts speaking before the sentence is finished.

Streaming TTSEnglish + हिंदी24 kHzNamed voices
Reso 1/v1.0STREAMING TTS

Warm, conversational male

0:00 / 0:07

Your appointment is confirmed for tomorrow at eleven. I've sent the details to your phone.

24 kHz·~150 ms to first audio·En + हिं
Capabilities

Built for live conversation.

01

Streaming, not batch

Reso 1 emits audio in chunks as it reads, so the first phoneme lands in milliseconds. Built for live phone calls, not file rendering.

02

Bilingual, with code-switching

Fluent English and Hindi in one voice, including the Hinglish that real Indian conversations are made of, switched mid-sentence without a seam.

03

Expressive control

Pace, emphasis, and intonation respond to punctuation and markup, so a confirmation sounds reassuring and an alert sounds urgent.

04

Consistent voices

Each named voice holds its identity across languages and a full call, no drift, no uncanny resets between turns.

Datasheet

The numbers.

Figures from internal evaluation. Final published metrics may vary by deployment.

Time to first audio~150msStreaming starts speaking before the sentence is parsed.
Naturalness4.4MOSMean opinion score from native-listener evaluation.
Audio24kHzStudio-grade output, chunked for low-latency streaming.
Faster than realtime5×One second of speech synthesised in ~200 ms on a single GPU.
Voices

Four voices, two languages, one model.

Each voice holds the same identity across English and Hindi. Press play to hear it, or ask us about a custom voice tuned to your brand.

Listen in

Aarav

Warm, conversational male

Your appointment is confirmed for tomorrow at eleven. I've sent the details to your phone.

Meera

Bright, expressive female

That's a great question. Let me pull up your account and we'll sort it out right now.

Kabir

Calm, authoritative male

For security, I'll send a one-time code to your registered number. Please read it back to me.

Isha

Crisp, neutral female

Thanks for waiting. Your refund of one thousand two hundred rupees is on its way.

The roadmap

Reso is a family.

Reso 1 is the first model out the door. Speech recognition and more Indian languages are training behind it, on the same streaming architecture.

Shipping

Reso 1 — TTS

Streaming text-to-speech in English and Hindi, with named voices.

Available to partners
In training

Reso ASR

Streaming speech recognition tuned for Indian accents and telephony audio.

Early access soon
Research

More Indian languages

Marathi, Tamil, Bengali, and Telugu, on the same streaming architecture.

On the roadmap
Request access

Put Reso 1 to work on your calls.

Tell us about your use case. We will set up Reso 1 and the products on your numbers, with India data residency and voices tuned to your brand.