Kontho (কণ্ঠ) — AI Voice Studio
Apnar kotha, je kono kontho-e, je kono bhashay. WhatsApp voice message er jonno.
Ki ki kore
- Meye ↔ Chele kontho (female→male, male→female), plus Baccha, Boyosko, Robot, Cartoon.
- Onno bhashay — Bangla, Hindi, English, Urdu, Arabic, Tamil, Telugu, Marathi, Gujarati, Nepali, Spanish, French, German, Russian, Chinese, Japanese.
- Tone — Swabhabik, Khushi, Shanto, Dukkhi, Uttejito, Ragi.
- Native pronunciation — Microsoft neural voice (prottek bhashar nijer male + female voice).
- Instant — kontho bodol phone-ei hoy, 1 minute audio ~1-3 second.
- WhatsApp — "WhatsApp-e pathan" button e share sheet khole, MP3 file jay.
- Khoroch ₹0 — sob free/open-source.
Kivabe kaj kore (architecture)
| Kaj | Ki use hoy | Khoroch |
|---|
| Kontho bodol (male/female/child/robot) | Nijer lekha TD-PSOLA engine (pitch + formant shift), phone-er browser-e | ₹0, internet lage na |
| Bola kotha → lekha | Browser Speech Recognition (Chrome), na hole Whisper (server) | ₹0 |
| Onubad | Google Translate free endpoint | ₹0 |
| Sundor AI kontho + onno bhasha | edge-tts (Microsoft neural voices) on free Vercel function (api/) | ₹0 |
| App hosting | Vercel / GitHub Pages (PWA, phone-e install kora jay) | ₹0 |
Folder
app/ — puro app (index.html, sw.js, manifest). Jekono free static host-e rakhun.server/ — free voice server (Hugging Face Space, Docker).src/ — source (dsp.js = voice engine, body.html = UI). node build.js diye app banay.
Sobcheye sohoj: Vercel e free deploy (AI voice soho)
- https://vercel.com/signup → "Continue with GitHub".
- "Import" → KANTHO repo → "Deploy".
- Je link pabe (jemon
https://kantho.vercel.app) seta-i full app. AI voice server (/api) eki sathe chole.
Alternative: Hugging Face server (5 minute)
- https://huggingface.co e free account khulun.
- New Space → naam
kontho → SDK: Docker → Blank → Public → Create.
server/ folder er 4 ta file (app.py, requirements.txt, Dockerfile, README.md) upload korun (Files → Add file → Upload).- 3-5 minute build hobe. Tarpor link hobe
https://<apnar-naam>-kontho.hf.space.
- App e ⚙ Settings → oi link din → "Server test" ✅.
Note: free Space 48 ghonta kono use na hole ghumiye pore; prothom request e ~30 second lagte pare.
Porer dhap (quality upgrade)
- RVC voice model: ekdom asol manush er moto meye/chele kontho (GPU lage; free Colab/HF ZeroGPU diye try kora jay).
- Nijer kontho onno bhashay (voice cloning, e.g. OpenVoice / XTTS) — apnar nijer golay Hindi/English.
- WhatsApp bot: voice note forward korlei bodle ferot pathabe (WhatsApp Business API — free tier e limited).
- Android app (PWA → TWA) Play Store e.
