Text to Speech Playground
Carnot Research — Accelerated Indic Speech & Translation AI Platform
High-throughput, ultra-low latency local GPU inference powered by NVIDIA L4 for 22+ Indic languages and full multi-speaker voice synthesis. Built for enterprise and government on-premise AI deployments.
Featured GPU Models
ai4bharat/indic-parler-tts
Multilingual, multi-speaker voice synthesis supporting natural voice controls, pacing, and 24 language locales.
ai4bharat/indic-conformer-600m
High-accuracy streaming Automatic Speech Recognition with low latency CTC decoding and VAD silence detection.
ai4bharat/indictrans2-en-indic-1B
State-of-the-art English → 22 Indic Languages Neural Machine Translation in FP16 precision.
Model Catalogue & AI Engine Profiles
Explore all 8 optimized Carnot local models running live on NVIDIA L4 GPU hardware.
ai4bharat/indic-parler-tts
Fully expressive 1.1B parameter neural TTS model trained on 24 Indic language locales. Supports speaker gender, emotion control, and variable pacing.
parler-tts/parler-tts-mini-v1
Lightweight 600M parameter streaming TTS engine optimized for real-time conversational agents and low latency audio output.
ai4bharat/indic-conformer-600m-multilingual
600M parameter multilingual ASR model supporting 22 Indic languages with streaming WebSocket chunking and VAD silence filtering.
distil-whisper/distil-large-v3
Distilled transformer ASR model delivering 6x faster inference speed than standard Whisper Large with sub-1% word error rate degradation.
ai4bharat/vakgyata-base-multilingual
Compact 300M parameter speech recognition engine tuned specifically for domain-specific vocabulary and noisy background environments.
ai4bharat/indictrans2-en-indic-1B
Encoder-decoder machine translation model translating English source text into 22 Indic target languages with FLORES codes.
ai4bharat/indictrans2-indic-en-1B
High-speed NMT model translating from any of 22 Indic languages into fluent English with sub-200ms latency.
ai4bharat/indictrans2-indic-indic-1B
Direct Indic-to-Indic language translation model (e.g. Hindi to Malayalam, Tamil to Telugu) without pivot English bottleneck.
// Generated code loading...{}
Drop audio file here or click to browse
Supports WAV, MP3, FLAC, OGG (Max 25MB)// STT Code...{}
// Translation Code...{}
API Keys
Your API keys grant access to Carnot Research STT, TTS, and Translation endpoints.
| Key Name | API Key Token | Created | Status | Actions |
|---|
GPU Hardware Telemetry & Microservices Status
Live health metrics for NVIDIA L4 GPU and active background processes.
Microservices Gateway Matrix
| Service Name | Protocol / Port | Active GPU Model | Memory Allocation | Status |
|---|---|---|---|---|
| Carnot Speech-to-Text | ws://127.0.0.1:8091 | ai4bharat/indic-conformer-600m |
264 MiB VRAM | Active (CUDA 12.4) |
| Carnot Text-to-Speech | ws://127.0.0.1:8092 | ai4bharat/indic-parler-tts |
2,438 MiB VRAM | Active (CUDA 12.4) |
| Carnot Translation Service | http://127.0.0.1:8000 | indictrans2-1B (EN-Indic & Indic-EN) |
4,394 MiB VRAM | Active (CUDA 12.4) |
| Carnot API Gateway Portal | http://127.0.0.1:3000 | Express API Gateway | 120 MiB RAM | Active |
Carnot Research API Reference & Developer Integration Guide
Comprehensive HTTP & WebSocket protocol specifications for STT, TTS, and Translation endpoints.
Text-to-Speech Endpoint (`POST /v1/tts/generate`)
Synthesizes input text into high-fidelity 44.1kHz audio in base64 format or WAV audio stream.
POST /v1/tts/generate
Header: Authorization: Bearer vx_live_...
Body:
{
"text": "नमस्ते",
"target_language": "hi-IN",
"voice": "shubh",
"pace": 1.0,
"sample_rate": 44100
}Speech-to-Text Endpoint (`POST /v1/stt/transcribe`)
Transcribes uploaded WAV/MP3 audio file or raw base64 PCM stream into text with language auto-detect.
POST /v1/stt/transcribe
Header: Authorization: Bearer vx_live_...
FormData:
file:
language_code: "hi-IN" Translation Endpoint (`POST /v1/translate`)
Translates source text between 22 Indic languages and English using IndicTrans2 FP16 on GPU.
POST /v1/translate
Header: Authorization: Bearer vx_live_...
Body:
{
"text": "Hello, welcome.",
"source_language": "English",
"target_language": "Hindi"
}