KanoAI: The Gujarati Language & Intelligence Suite

📄 Reference (Light) Manual Inkscape

⚡ Auto-Generated (Bold) Medial Ridge Snapped

🗣️ ગુજરાતી Text-to-Speech (TTS) Studio

State-of-the-Art Gujarati Neural Voice Generation • AI4Bharat IndicF5 (Flow-Matching) & Indic-TTS / Indic-Speak (Multi-Speaker)

📊 Side-by-Side Model Comparison →

Checking Server...

Gujarati Text Input (ગુજરાતી લખાણ) 0 chars • 0 words

⚡ Quick Sample Sentences:

Voice & Conditioning Settings

TTS Engine / Architecture: 100% Offline VITS

End-to-end Variational Inference TTS (facebook/mms-tts-guj) running 100% locally on CPU/MPS with zero cloud dependency.

📊 Listen to side-by-side audio comparison of all 8 models & 25 flavours →

Style / Voice Conditioning Preset: Natural Language Voice Prompt: Parler API Endpoint:

Cloud ZeroGPU backend on Hugging Face Spaces.

Device Native Gujarati Voice: Voice Pitch:

⚡ 100% Client-Side. Runs directly inside your browser via OS hardware speech synthesis (Google / Apple gu-IN) with 0ms server latency.

Bhashini Voice Profile: Bhashini API Endpoint:

⚡ Operates 100% locally and offline on your machine with zero cloud dependency.

Google Cloud Voice Model: Google Cloud API Endpoint:

Cloud Text-to-Speech API with Linear16 studio audio. Requires Google Cloud API Key.

Google Cloud API Key:

Model Checkpoint: MMS-TTS API Endpoint:

100% Offline CPU/MPS VITS inference via local Unified Backend (port 8000).

Piper Neural Voice Model: Piper TTS API Endpoint:

Sub-100ms ultra-fast ONNX neural runtime via local Unified Backend (port 8000).

Ultra-fast neural text-to-speech running locally via ONNX Runtime with sub-100ms CPU inference.

eSpeak-NG Phoneme Table: Voice Timbre & Filter Variant:

💡 Notice: eSpeak-NG is a 1990s rule-based formant acoustic synthesizer (<2 MB, zero neural weights), so it has an inherently synthetic/robotic timbre. For realistic, authentic human speech, select Meta MMS-TTS or Piper TTS.

Voice / Speaker Profile:

Warm, natural, and expressive female voice with clear Gujarati diction.

Indic-TTS API Endpoint:

Local server command: python python/tts/run_local_indic_tts.py --port 7861

Reference Audio:

Voice conditioning from native Kano pronunciation dataset (Kakko 1_ka.mp3).

IndicF5 API Endpoint:

Local server command: python python/tts/run_local_indic_f5.py --port 7865

Speech Pacing & Speed:

📸 ગુજરાતી ઓપ્ટિકલ કેરેક્ટર રેકગ્નિશન (Gujarati OCR Studio)

AI4Bharat & Bhashini IndicPhotoOCR, Gujarati TrOCR & GujaratiHCR • Neural Vision for Scene, Printed & Handwritten Documents

Checking API Server...

📥

Drag & Drop Gujarati Image Here, Paste (Ctrl+V), or Click to Browse

Supports JPG, PNG, WEBP • Printed pages, Signboards, Handwritten Notes

📚 Benchmark Preset Samples:

Printed Text: Storybook page with headers and paragraphs. • Recommended Engine: Bhashini IndicPhotoOCR (Fast)

OCR Engine:

Bhashini-IITJ IndicPhotoOCR (Offline Mode): Local scene text detection (TextBPN++ contour pipeline) & Gujarati recognition on port 7860/8000.

IndicPhotoOCR Mode:

IndicPhotoOCR Server URL: ● Active (Port 7860 / 8000)

TrOCR Mode:

TrOCR Local Server URL: ● Online (7862)

Language Identifier:

Hugging Face Token (Optional): Get Token ↗