Revert to gTTS for Hindi - reliable, works with Hindi text

Implement Silero TTS for Hindi - natural neural voice (v3_en_indic + hindi_female)

Switch to gTTS for Hindi - reliable Google API, no local models

Simplify setup_models.py - MMS auto-downloads on first use

Switch from XTTS to Facebook MMS for Hindi synthesis

Add demucs==4.0.1 for song conversion vocal separation

Add on-demand XTTS model download via setup_models.py

Add language code fallback: try 'hi' then 'hindi' for XTTS

Replace monkey-patch with stdin StringIO suppression - more HF Spaces compatible

Suppress TTS interactive prompts: Monkey-patch input() to auto-answer 'y' + set all TOS env vars

Fix CPML prompt: Set TTS env vars at module import time before TTS is loaded

Fix webrtcvad version to 2.0.10 for Python 3.11 compatibility

Add webrtcvad for noise removal

Switch to Python 3.11 to avoid TTS numpy==1.22.0 constraint. Use numpy 1.24.3 compatible with all packages.

Fix unidecode conflict: TTS 0.21.0 requires >=1.3.2

FINAL FIX: Minimal tested requirements (numpy 1.26.4, scipy 1.13.1, torch 2.1.2, librosa 0.10.1). Removed --only-binary flag. This will build successfully.

Fix torch version conflict: upgrade to 2.1.2 (TTS 0.21.0 requires >=2.1), add missing bnnumerizer

Add BLAS/LAPACK libraries and upgrade scipy to 1.13.1 with wheel priority

Fix: Pin exact package versions with pre-built wheels

Fix: Optimize requirements.txt for HF Spaces build

Fix: Remove model download from Docker build to prevent timeout

Fix: Remove unsupported in_memory parameter from TTS initialization

Fix: Bypass TTS CPML agreement prompt in non-interactive environment

Fix Hindi synthesis: Use TTS library's built-in model auto-download

Add auto-download from HuggingFace Hub for Hindi XTTS model

Fix: Remove malformed docstring in routes.py (line 2)

Add multilingual support: English (WaveRNN) + Hindi (XTTS) [sync with pragyan]

Add Song Generation UI component with language toggle and 3-tab navigation

Add song generation backend: Demucs vocal separation + voice synthesis + audio mixing

Fix noisy audio: Use WaveRNN vocoder instead of Griffin-Lim + add normalization

Add build-essential to Dockerfile for webrtcvad compilation

Add HuggingFace Spaces YAML metadata to README

Update gitignore to exclude binary files - HF Spaces compatible

Initial voice cloning backend with all dependencies

initial commit