Spaces:
Running on Zero
A newer version of the Gradio SDK is available: 6.25.0
title: Urdu S2S MVP
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 6.8.0
pinned: false
license: other
short_description: Urdu speech-to-speech and text-to-speech demo
Urdu S2S MVP
This Space runs the current Urdu speech experience: a Pakistani Urdu voice assistant demo for conversational speech-to-speech and text-to-speech testing.
Use Speech to speech to record or upload audio and receive a spoken Urdu response. Use Text to speech to type Urdu, Roman Urdu, or Devanagari text and generate speech.
About the Model
Urdu S2S is an experimental voice model focused on natural spoken Pakistani Urdu, short assistant-style replies, and a warm female assistant voice. It is intended for interactive product testing rather than long-form narration.
Architecture
Urdu S2S uses a modular speech pipeline with dedicated components for speech recognition, Urdu response generation, pronunciation-aware speech preparation, and neural speech synthesis. The user-facing experience is a single speech-to-speech model demo, while the underlying design lets each part improve independently as stronger Urdu speech models become available.
At a high level, incoming audio is normalized and transcribed, the dialogue layer produces a concise Urdu assistant response, a pronunciation layer prepares the response for stable speech generation, and the synthesis layer renders the final spoken voice.
The current demo is tuned for:
- Urdu speech-to-speech conversations
- Urdu, Roman Urdu, and Devanagari text-to-speech
- Short support, scheduling, helpdesk, and everyday assistant prompts
- Urdu-first wording and pronunciation preferences
- Neutral user addressing without assuming the speaker's gender
How to Test
For best results, use short recordings of 2 to 10 seconds with clear speech. Try prompts such as greetings, clarification requests, simple account-support questions, scheduling drafts, travel questions, or customer-service style requests.
Current Limits
This is an MVP. Very noisy audio, long recordings, dense factual questions, unusual names, and heavy code-switching may still create transcription or pronunciation errors. The first generation after the Space wakes up may be slower while the speech model loads.
Deployment
This Space is configured for interactive demo usage with GPU-backed speech generation.